Introduction

Optical microresonators1 enable or enhance a diverse set of functions in integrated optical systems, spanning filtering and routing2, switching and modulation3, control of light emission4, photodetection and sensing. Low-volume optical modes are advantageous for a variety of reasons in all of these applications; examples include smaller footprint for dense integration, greater sensitivity to local index perturbations for sensors, greater energy efficiency in tuning for modulation, lower capacitance devices, stronger nonlinear interactions and coupling to light emitters such as quantum dots and atoms. As a result, due to their ability to support very high-quality-factor (Q) resonant optical modes with mode volumes on the order of (λ/n)3 or less, PhCs have attracted interest from a broad range of fields in recent years5,6,7,8,9. Much progress has been made in the optimization and fabrication of such structures, but due to the ~100-nm feature sizes required for structures with photonic bandgaps in the optical or near-infrared, fabrication typically relies on electron-beam lithography. This has limited the practical use of PhC structures, since it precludes practical integration with large-scale photolithographically defined systems and also makes challenging the creation of complex structures requiring multiple masks9.

Although the possibility for photolithographically defined quasi-2D PhCs has been explored in the past10,11, high-Q resonators proved difficult to achieve. Passive optical cavities with measured quality factors (and extracted intrinsic Qs) of approximately 3,000 (10,000)12 and 2,200 (on the order of 100,000)13 have been designed and fabricated in SOI CMOS processes, but the potential for reliably creating active electro-optical devices in electrically-contacted PhC cavities capable of Qs exceeding 105 in a scalable process has remained unexplored.

In this work, we demonstrate PhC microresonators with radiation-loss limited Qs well in excess of 105 and measured material loss-limited Qs up to 58,000, fabricated photolithographically within a scaled CMOS process in the same patternable polycrystalline silicon (pSi) layer as used for transistor gates. A 1.2 μm thick, patterned oxide beneath optical devices eliminates the need for undercuts or any post-fabrication processing. We introduce also a laterally electrically-contacted design for a quasi-1D resonator utilizing a patterned, partially etched silicon layer and demonstrate electro-optic functionality in the form of defect-state photodetection in a resonant p-i-n structure which allows efficient photodetection at 1550 nm wavelengths in an all-silicon CMOS device. These results indicate the potential for high-Q PC resonators to achieve both passive and active optical functionalities in scaled photonic systems, in a way that can be straightforwardly integrated with sophisticated electronics14. The work indicates the practicality of silicon PhC microcavity devices for application in the near term for silicon photonic interconnects15,16,17 and more speculatively in the long term for integrated quantum photonic systems18.

Results

A schematic chip cross section around the pSi layer in the process used is shown in Fig. 1(a), showing p- and n-mos transistors and doping profiles in the crystalline silicon substrate. Passive optical structures can be formed within the undoped pSi used as the transistor gates and losses down to 6 dB/cm at 1550 have been previously reported in waveguides formed in this layer19. A partial etch step allows thin wings to sit adjacent to the waveguide cores, which can be doped and electrically contacted with metal vias which access three patternable copper wiring layers. A conformal silicon-nitride (n ≈ 2.0) layer encloses the pSi and fills the holes in these designs which are otherwise SiO2-clad, resulting in a lower index contrast than the Si/air structures which have achieved the highest Q/V ratios.

Figure 1
figure 1

(a) Photograph of full wafer, on which a single reticle (inset) was repeated. The location of the PhC devices discussed here is labeled within the inset. (b) Schematic chip cross-section (not to scale), illustrating (from left to right) doping profiles around nmos and pmos transistors with metal vias and copper wiring, intrinsic pSi waveguide structures separated from the silicon substrate by a 1.2 μm-thick Deep Trench Isolation (DTI) deposited oxide and a doped and contacted electro-optic structure. (c) Detailed schematic including dimensions near optical cores. (d) Transformation of supplied (desired) pattern to mask design after optical proximity correction and SEM images of 2D lattice of fabricated holes in pSi.

Full size image

A 193 nm photolithography process with 0.68 numerical aperture was used to define features in the pSi. 100-nm features transferred to a photoresist using such illumination can suffer significant optical proximity effects10 and here an optical proximity correction (OPC) algorithm partially compensated for such effects in definition of the circular holes; Fig. 1(d) shows the OPC-generated mask pattern for a triangular lattice of circular holes and an example SEM image of a 2D lattice fabricated, with 375 nm lattice constant. From a set of SEM images on devices with a variety of lattice constants (a) and hole radii, an approximately linear relation between specified and achieved hole radii was found that was independent of lattice constant and was used to compensate for uniform lithographic bias in future designs in the same process. Holes with radii down to about 55 nm formed reliably.

A variety of previously studied PC microcavity designs would be suitable for integration here. Minimization of radiative loss into the cladding motivates choice of cavities with smooth, ideally Gaussian envelopes5,20,21 and quasi-1D structures, relying on photonic bandgap confinement in only one direction, are particularly robust in moderate index-contrast environments22. The bulk of our devices relied on such cavities.

Quasi-1D resonators for operation near 1550 nm using a linear array of holes in a 225 nm-height pSi waveguide of width ≈ 450 nm were designed with a lattice constant a near 330 nm, based on the hole tapering scheme developed in23. Around the cavity center to either side, hole radii decrease linearly from r = 0.33a to r = 0.28a; a set of “mirror” holes, all with r = 0.28a are added to either side, the number of which controls the strength of coupling to the feeding bus waveguides24. In the dielectric environment here, these resonators achieve intrinsic (that is, with no coupling to the feeding waveguides) radiative Qs of over 2 million, as calculated with finite-difference time-domain (FDTD) simulations25 without material loss.

To allow efficient electrical contact to the resonator without introducing large optical loss, thin ≈ 50 nm-thick pSi “wings” were placed adjacent to the cavity. However, continuous slabs, in conjunction with the SiN liner layers, had effective indices large enough to couple to some Fourier components of the resonant mode, resulting in optical leakage, which could be reduced by patterning the wings with a 2D lattice of triangular low index holes; this had the effect also of reducing the resonant mode field's evanescent decay into the wings, allowing doped regions to be brought closer to the cavity (see Supplementary Information). This patterning was done with a lattice of r = 0.3a holes here. An SEM image of a resulting structure along with the FDTD-calculated resonant electric field profile is shown in Fig. 2(b). The intrinsic radiative Q of the contacted structure was calculated to be ≈ 300,000 and the mode volume 0.8(λ/n)3.

Figure 2
figure 2

(a) Optical micrograph of a single cavity device, with grating couplers and tapers for input/output light coupling. (b) SEM image of pSi in a fabricated PhC device, with FDTD-calculated mode profile overlaid. (c) and (d) Measured transmission spectra through cavities with resonant wavelengths λ0 ≈ 1512 (a = 320 nm, w = 450 nm) and λ0 ≈ 1549 nm (a = 330 nm, w = 470 nm), respectively. Curves and fits for cavities with 8, 12 and 18 pairs of mirror holes are shown for each (black, red and blue curves), with (e) peak resonant transmissions fit to theoretical predictions as function of total quality factor, allowing extraction of intrinsic Qs of 58,000 (black curve for λ0 = 1512 nm cavity) and 51,000 (red curve, λ0 = 1549 nm).

Full size image

Light is coupled to independent devices via grating couplers from fibers oriented nearly normal to the chip surface; adiabatic tapers couple to the ≈ 500 nm-wide waveguides, as shown in the optical micrograph of Fig. 2(a). Transmission spectra through cavities with resonant wavelengths λ0 = 1510 nm and λ0 = 1540 nm, with three different waveguide coupling strengths (8, 12 and 18 pairs of mirror holes) are shown in Fig. 2(c) and (d), along with theoretical fits to Fano profiles26 which allowed determination of each cavity's loaded quality factor. The peak transmissions of each cavity, measured and normalized to the grating coupler transmission loss calibrated separately, closely follow the theoretical prediction of a standard coupled mode analysis Tpk = (Qtot/QWG)227. Here the total measured quality factor is related to the waveguide coupling strength via the relation , where Q0 is the cavity's intrinsic loss rate (due to material and radiative loss) and QWG is the quality factor associated only with the cavity loss rate into the bus WG modes. As shown in Fig. 2(e), the peak transmissions and measured total quality factors are closely fit by the theoretical prediction for intrinsic quality factors Q0 = 58,000 and 51,000.

These loss rates are thus clearly limited not by the device design, but predominantly from material loss in the pSi and fabrication imperfections. Characterization of propagation loss in rectangular waveguides on the same wafer gives insight into the role of bulk material loss as well as sidewall roughness. The intensity loss coefficient of the bulk material in these samples is measured to be approximately α ≈ 4 dB/cm ≈ 0.92 cm−1, corresponding to a temporal photon loss rate of γ = αc/ngr ≈ 8 GHz, where ngr is the waveguide group index. This results in a material loss-associated Q of Qmat = ω/γ ≈ 154,000. However, sidewall roughness makes a significant contribution to loss in these devices as well. Rectangular waveguides of 400 nm width, in which the mode profile has significant overlap with the sidewalls, exhibited propagation losses of about 8.5 dB/cm at the wavelengths of interest. Although the overlap with the sidewalls in the PhC cavities can be expected to be only roughly that of the fundamental mode in the 400 nm rectangular waveguide, this measured loss would correspond to a total Qmat + roughness ≈ 77, 000.

Variance in fabricated hole sizes could play a role as well; scattering loss has been resulted to scale approximately as , where σr represents the RMS radius deviation28,29 and A represents a design and index-contrast-dependent parameter. Scanning electron microscope characterization of fabricated samples indicates σr ≈ 2.0 nm and FDTD simulations incorporating random disorder in the hole sizes with Gaussian distribution indicate A ≈ 0.1 (see supplementary information), resulting in a Qdis ≈ 280,000. Together, these loss rates are consistent with the observed intrinsic Qs. Use of higher NA photolithography is known to allow lower sidewall roughness and will likely improve disorder in the PhC features as well, which presents a clear path to reducing the impact of fabrication imperfections on these loss rates.

A significant fraction of the material loss (the dominant loss mechanism in these devices as discussed above) results from absorption events involving pSi grain boundary defect electronic states at energies within the bandgap19,30. Since such absorption events generate free carriers, a significant photocurrent may be expected to flow in the contacted diode structures, which should be observable under reverse bias. Such photodiode behavior has indeed been previously observed in pSi ridge waveguide ring resonator structures31; in crystalline silicon, such detection has been observed in samples with defects introduced by ion implantation with high responsivities of 0.7 A/W32 and even with avalanche gain allowing photodiode operation with 10 A/W operation and >35 GHz bandwidths33. Photodetection has been reported also in defects inherent at the Si/SiO2 interfaces34 and defect photogeneration mechanisms have been applied to resonant photodetectors as well35,36.

A fabricated 1D PhC structure with lateral electrical contacts is shown in Fig. 3(a), with p+ and n+ doped regions shaded in red and blue. Variants were tested with doped region spacings of wi = 1.6 μm (structure and results shown in Fig. 3), as well as 1.0 and 2.2 μm. The cavity is coupled to a feeding waveguide from only one side and ideally the loaded Q including coupling into the bus waveguide mode would be half that of the intrinsic Q, to ensure that no light is reflected into the bus, i.e. that the cavity is critically coupled27. In these devices, material loss in the pSi was higher due to additional processing steps and devices had measured intrinsic Qs of approximately 22,000 with no p+/n+ regions nearby; the loaded Q of the resonator devices with wi = 1.0, 1.6 and 2.2 μm were approximately 4,300, 8,000 and 9,300, respectively, indicating closeness to critical coupling for the larger intrinsic region widths and the effect of free-carrier absorption for the device with lowest wi.

Figure 3
figure 3

(a) SEM image of detector structure with one optical input, with p+ and n+ implanted regions shaded. Measurements are presented in this figure for a device with intrinsic region width wi = 1.6 μm, as labeled. (b) IV characteristics without illumination and with 3 μW in-waveguide power at cavity resonance wavelength. (c) Photocurrent spectra at reverse bias of 22 V for powers logarithmically spaced between 1.6 and 120 μW input optical power and (d) Peak QE vs. voltage and power for resonant pSi PhC defect-state photodetector.

Full size image

Dark current in the wi = 1.6 μm structure is below 500 pA at reverse biases up to 30 V (as shown in Fig. 3(b)), with large separation from the photocurrent when illuminated on resonance (near 1538 nm in this device) over a large voltage range. Photocurrent spectra obtained at a reverse bias of 22 V for input powers between 1 and 100 μW are plotted in Fig. 3(c); the carrier photogeneration rate is proportional to stored cavity energy and the spectra are thus Lorentzian at low input powers and exhibit bistability resulting from the thermo-optic effect37 at higher powers. The peak current relates directly to the onresonance quantum efficiency (QE), plotted over a range of input powers and biases in Fig. 3(d); QE here is defined as number of charge carriers extracted per photon incident on the cavity, calculated from Ipkhν/Pinq, where Ipk is the peak resonant photocurrent and Pin is the optical power input after the grating coupler. A strong voltage dependence is observed and large voltages of at least 15 V are required to achieve QEs above 10%, likely due to inefficient extraction under low fields due to recombination in the 1.6 μm-wide intrinsic region. The power dependence of the peak QE appears similar to that previously reported in resonant pSi photodetectors31 and could be due to density-dependent recombination rates; this effect must be stronger than any contribution from two-photon absorption at the higher powers in the range of powers studied here, since no increase in QE at high optical powers is observed. We applied a maximum of 30 V bias to the device and observed a peak QE of 28% (0.35 A/W). At voltages in this range, material properties of the pSi appear to have been affected in a way which resulted in permanent (stable at least for many days) changes in the device response; the data presented here was verified to be stable and we briefly discuss some of the changes observed in the Supplementary Information.

Geometry also strongly influenced the modulation response of the photodetectors. Devices with narrower intrinsic regions had significantly faster 3-dB bandwidths; a modulation response curve with an empirical fit is shown in Fig. 4(a) for a device with a 1 μm-wide intrinsic region with -30 V applied (a DC QE of 25% and 30 nA dark current were measured in this device and bias), indicating a 2.9 GHz 3-dB roll-off. Oscillations around the fit are repeatable and due primarily to impedance mismatch between the device's large series resistance and the 50 Ω load. The measured bandwidth increases strongly with increasing reverse bias and is lower in devices with larger intrinsic region widths (the three curves in Fig. 4(b)). The device's RC limit is significantly higher than the measured roll-offs owing to its extremely small capacitance (), as is the RC limit owing to the load 50 Ω resistance and the contact pad capacitance; the measured frequency responses are, however, plausibly explained in terms of the limit imposed by the transit time through the intrinsic region, which increases with intrinsic region width and varies inversely with mobility. Both electron and hole mobilities in pSi thin films are known to decrease with film thickness38, consistent with the strong dependence on the width of the 50-nm partial etched wings the carriers have to traverse in these devices.

Figure 4
figure 4

(a) Modulation response of detector with 1 μm intrinsic region, under 30 V reverse bias with an empirical fit showing 2.9 GHz 3-dB frequency. DC quantum efficiency at this operating point was 25% (responsivity 0.31 A/W), with a dark current of 30 nA. (b) Fit 3-dB frequencies for 3 different devices with identical optical design, but different intrinsic region widths (wi, as labeled; the middle curve shows results from the same device as in Fig. 3) as a function of applied reverse voltage. Lines are guides to the eye.

Full size image

Discussion

Simple improvements to the electrical design of the photodetectors presented above should allow faster response, along with higher quantum efficiencies at lower applied biases. Incorporating mid-level doped regions in addition to the p+ and n+ used in these structures so as to form a p+/p/i/n/n+ structure with p and n regions extending closer to the device center would allow lower wis without introducing excessive optical loss and therefore higher transit time-limited bandwidths (inversely proportional to wi); thicker contact wings would also increase bandwidth by increasing mobility in the wings. Both of these modifications would also be expected to reduce the applied voltages required for carrier extraction by reducing recombination in the intrinsic region.

The photodetector results presented above constitute a first realization of electrically active PhC devices in a CMOS environment, but together with the possibility for increased bandwidth and reduced operating voltage also indicate promise generally for resonant subbandgap photodetection with sub-nA dark currents and low capacitance in this environment. Use of higher NA photolithography in more modern processes should also allow for lower roughness and disorder-induced loss and higher intrinsic Qs in PhC devices generally.

Our work demonstrates that pSi PhC microresonators with quality factors of up to 60,000 in passive devices can be designed and fabricated within a scaled CMOS foundry in the same layer as is used for transistor gates, with potential for significant improvement with process and material quality. The contacted designs and the photodetector device presented here also indicate the potential for active wavelength-scale electro-optic devices. In the context of integrated Si photonic interconnects, we expect these results could impact design considerations primarily for low-capacitance resonant photodetectors (both linear and nonlinear) and high-efficiency modulators. Quantum photonic applications relying on such resonators may eventually benefit as well from the ability to produce them in scaled processes, integrated tightly with control electronics and tuning mechanisms.

Methods

Cavity simulation and design

Design proceeded by first calculating band structures of 1D PhC waveguides39 to approximately locate parameters for the desired resonant frequency, after which FDTD simulation in MEEP25 of the full design allowed for quality factor calculation and optimization. Band structures of the 2D contact slab, as shown in the supplementary information, were calculated with the same method as for the 1D cavity waveguides. Designed devices were laid out within Cadence Design Systems' Virtuoso, a common VLSI layout tool40.

Device characterization

An HP 8164A lightwave measurement system was used for spectral measurements; the laser linewidth (and hence spectral resolution) is < 1 MHz. An Agilent 4156C semiconductor parameter analyzer was used for DC biasing and current measurement. For modulation response measurements, the output of a microwave synthesizer was modulated onto the input light with a lithium niobate modulator and the RF power in the device photocurrent measured with a microwave spectrum analyzer. Consideration of the RF power measured in the PhC device relative to that under the same measurement performed on a commercial photodiode with a frequency response known to be flat (Discovery Semiconductor DSC20S) for frequencies up to 20 GHz allows for elimination of any frequency response from components in the system other than the PhC device.