3.1.

Integrated Interferometers

Interferometers are historically established as highly sensitive sensors based on the phase-sensitive detection of minute optical path differences; a striking recent example is the detection of gravitational waves.¹¹ Interferometric PIC-based sensors, like their traditional free-space counterparts, split light from one waveguide into two separate arms, a reference arm and a sensing arm, functionalized for analyte capture. If a target analyte is present in a sample, it will bind to receptors on the sensing arm, and, as a result, light passing through the sensing arm will be phase shifted relative to the reference light. The interference pattern formed when light from both arms is recombined can then be measured to determine the relative optical phase delay and calculate the amount of analyte present. Two standard configurations exist for interferometers¹² integrated on a chip: the Mach–Zehnder interferometer (MZI)¹³ [Fig. 3(a)] and the Young interferometer (YI) [Fig. 3(b)], which is similar but out-couples the light in both arms allowing the light to interfere in free space, instead of using a junction on-chip.

Fig. 3

Interferometric PIC sensors. A standard (a) MZI (a) and (b) YI configurations. Light enters the interferometers on the left and is split into the sensing and reference arms. In the MZI case, the arms are recombined in a Y-junction and the interference measured in the output is used to quantify the presence of an analyte in the sensing arm. In the YI configuration, the light is not coupled, but is outcoupled from sensing and reference arms and interfered in free space to create interference pattern. The presence of an analyte and its concentration is identified based on the interference pattern, reproduced from Ref. 14.

Several demonstrations of MZI-based PIC biosensors are reported in the literature, with LODs reaching as low as $1\times {10}^{-7}\mathrm{RIU}$ or, equivalently, LOD of 10 pM of DNA.¹⁵ Various waveguide designs, such as slot and bimodal waveguides (BiMW), have also been explored to improve sensor accuracy. In the slot design, the sensing arm is split into two waveguides separated by a nanometer-scale, low-refractive index “slot” region for light confinement. This design greatly enhances the light–matter interaction yielding higher sensitivity to the refractive index of the analyte; an LOD of $5.4\times {10}^{-6}\mathrm{RIU}$ and $1\mathrm{pg}/\mathrm{mL}$ of streptavidin was recently demonstrated.¹⁶ The same slot design was also recently integrated into an LOC-type point-of-care medical device that used a multiplexed approach to detect specific microRNAs in human urine with an LOD of 1 nM.¹⁷ In the BiMW configuration, a step junction in a single-mode waveguide excites two transverse modes, in a single waveguide, which interact with the sample before being measured by a photodetector. The sensitivity level of the BiMW sensor is comparable to other integrated interferometers; LODs down to $5.9\times {10}^{-4}$ experimentally were recently demonstrated for bovine serum albumin (BSA), with a theoretical LOD limit of $2.5\times {10}^{-7}\mathrm{RIU}$.¹⁸

To the best of our knowledge, the most sensitive biochemical interferometric sensor is a YI device with two ${\mathrm{Ta}}_2{\mathrm{O}}_5$ slab waveguides, which achieved a sensitivity of $9\times {10}^{-9}\mathrm{RIU}$ and a surface coverage of $13{\mathrm{fg}/\mathrm{mm}}^2$ for both bulk solution and surface sensitivities to immunoglobulin G.¹⁹ However, the YI configuration requires off-chip detection and analysis, which may impact operationally relevant considerations for DoD applications.

3.2.

Ring-Resonator Sensors

Ring-resonator transducers are becoming more common for biochemical sensing applications due to their high sensitivity and potential for multiplexing; the detection of multiple analytes is vital for testing complex fluids (e.g., blood, saliva, and urine) in diagnostics, monitoring, and toxicity screening.²⁰ PIC ring-resonator sensors use at least one linear waveguide to couple light to a closed-loop waveguide (the ring resonator) and, through evanescent-wave coupling, excite resonant modes in the loop waveguide. On resonance, constructive interference is generated in the multiple roundtrips over the ring circumference; however, off resonance the transmission rapidly decreases. This resonance effect considerably enhances sensor sensitivity.²¹ Analytes captured on the functionalized surface of the ring resonator shift the resonant wavelength only slightly, but this change is detected as a relatively large decrease in output intensity at a specific wavelength [Fig. 4(a)].¹⁰

Fig. 4

Ring-resonator sensor. (a) Schematic of a ring-resonator sensor. The linear waveguide provides input light to the loop waveguide at the coupling zone and captures output light from the loop for measurement. The presence of analyte will shift the resonance wavelength within the loop, which is detected in the output spectra, adapted from Ref. 10. (b) Micrograph of a fabricated ring resonator, adapted from Ref. 22.

The number of revolutions light takes around the ring resonator yields an optical path length (OPL) orders of magnitude larger than the sensor’s physical footprint, which is generally described by the resonator’s quality ($Q$) factor.²¹ High $Q$-factors—considered to be at least 106—indicate low optical loss and correspondingly long photon lifetimes, which is translated into narrow linewidths and often high peak resolution (i.e., high sensitivity). $Q$-factors in the 106 range have been reported for resonators around 50 to $200\mu \mathrm{m}$ in circumference, which has an OPL equivalent to a linear waveguide measuring several centimeters. In practice, a $Q$-factor of 108 effectively means a molecule will be sampled more than 100,000 times, given by $L_{\mathrm{eff}}=\frac{Q\lambda }{2\pi n}$, where $L_{\mathrm{eff}}$ is the effective free-space path length, $Q$ is the $Q$-factor, $\lambda$;is the free-space wavelength, and $n$ is the refractive index of the resonator.²¹

Other important characteristics of resonators are the free spectral range (FSR) and the finesse. The FSR is the inverse of the round-trip time of an optical pulse around the resonator and defines the optical frequency range over which the resonator can be utilized. The finesse, which is determined by the resonator losses and in independent of the resonator circumference, is defined as the FSR divided by the full-width at half-maximum bandwidth of the resonance. Hence, $L_{\mathrm{eff}}$ can also be defined as the finesse times the ring circumference, which is a useful metric as the sensitivity generally scales with finesse for most sensing applications.

Therefore, despite the small physical size of the ring resonator, it can achieve higher sensitivities than straight waveguides while using orders of magnitude less surface area. Li and Fan²³ have demonstrated LOD as low as $2.5\times {10}^{-7}$ in RIU, or equivalently detected $1.6{\mathrm{pg}/\mathrm{mm}}^2$ of biotin, using a ring resonator with a diameter of just $70\mu \mathrm{m}$. Smaller waveguide sizes also facilitate the development of more complex, integrated devices. For example, Iqbal et al. demonstrated an arrayed device using 32 ring-resonator sensors to detect streptavidin diluted in BSA with an LOD of 60 fM. Furthermore, pushing the technology toward detection of analytes in complex environments, Iqbal et al.²⁴ demonstrated simultaneous detection of two DNA oligonucleotides. Isolating 8 of the resonators to use as control sensors, they separated the remaining 24 sensors into two groups and functionalized each for a specific oligonucleotide and successfully demonstrated multiplexed sensing. More recently, there have been demonstrations of functionalizing ring resonators with sorbent polymers to efficiently bind vapor-phase compounds, such as acetone and DMMP, and achieve LODs as low as a few ppb.²⁵

3.3.

Photonic Crystal-Based Sensors

PhC are structured matrices of materials that possess different dielectric constants, resulting in photonic bandgaps that determine the frequencies of light, which are reflected or transmitted by or through the substrate.²⁶ The simplest type of PhC structure is a perfect array of holes, which was in fact one of the earliest geometries for PhC biochemical sensing;²⁷ the PhC was designed to reflect a single wavelength of light and, upon analyte binding to immobilized receptors on the surface (e.g., immunoglobulins, streptavidin, and BSA), a change in the reflected light wavelength was measured with a reported detection limit of $500{\mathrm{ng}/\mathrm{mm}}^2$.²⁸

The introduction of defects into the PhC structure (i.e., shifting or removing holes, or changing hole diameters) results in the confinement of light to the defect with high $Q$-factors, enabling highly sensitive detection of extremely small sample volumes (e.g., a few $\mu {\mathrm{m}}^3$). The prospect of high $Q$-factors and small sample volumes has led to significant effort to explore the potential of PhCs for biochemical sensing, beyond the mirror geometry discussed above.²⁹ Increasingly complex structures, such as that shown in Fig. 5 where the use defects (i.e., the absence of holes) introduces a waveguide, have been demonstrated to confine sensing volumes and increase sensitivity. For example, slot waveguides have been integrated in PhCs to boost sensor performance, enabling a $Q$-factor of up to 50,000, a sensitivity of $1500\mathrm{nm}/\mathrm{RIU}$, and an LOD of $7.8\times {10}^{-6}\mathrm{RIU}$.³¹ While analyte detection was not demonstrated with the device in Ref. 31, similar PhC devices have been used to detect lung cancer cells down to $2\text{cells}/\mu \mathrm{L}$, to measure biotin–streptavidin binding (a common analyte/receptor pairing to demonstrate biochemical assay viability), and to probe the binding kinetics of immunoglobulin G proteins.³² More recently, PhC biosensor devices have been developed for detection of HIV,³³ biomarkers for iron deficiency anemia,³⁴ and glucose monitoring.³⁵

Fig. 5

A 2-D PhC waveguide adapted from Ref. 30. The top and bottom milled areas form a PhC with a photonic bandgap. Light is confined to defects which, here, is the region absent of holes, which forms the central waveguide. Introducing analytes near the PhC changes the RI and the cutoff frequencies of the bandgap.

3.4.

Summary and Comparison of Single-ID PIC-Based Sensor Characteristics

As with all technologies, specific features and designs of single-ID PIC-based sensor types carry a cost/benefit trade-off, particularly for operationally demanding DoD use cases. Interferometric devices look very promising as they are inexpensive to fabricate and are the most sensitive PIC sensors, achieving LODs as low as $9\times {10}^{-9}\mathrm{RIU}$; however, interferometric detectors with high sensitivities are physically large, provide enough functionalized surface area, and currently rely on off-chip excitation and detection schemes.

Ring resonators and PhCs have good miniaturization potential, a feature almost universally coveted for DoD biochemical sensing applications but also have challenges. As with all high-$Q$ resonant devices, both are susceptible to thermal drift, or a thermally induced change of the refractive index given by the thermo-optic coefficient.^36,37 The inherent high sensitivity of the resonant condition to refractive index makes thermal drift the primary detuning mechanism. Several methods have been identified to mitigate thermal drift, including making devices thermal using core and cladding materials with opposite signs in their thermo-optic constants, or using a reference resonator, which is identical to the sensing elements but placed in a separate fluidic channel. The latter method can be utilized for PhC-based sensors as well,³⁸ with more sophisticated methods encoding a reference resonance that is robust to environmental changes.³⁹ It is important to note that interferometric PIC-based sensors are less susceptible to temperature drift, which makes them a highly attractive platform for field-based sensing applications.

Still, ring resonator and PhC sensors both have advantages. Ring resonators can be easily integrated into multiloop detector circuits for multiplexed and LOC-type applications, which will likely be critical for fielded devices. PhCs, likely the least proven of the three PIC types for biochemical detection within integrated devices, are perhaps of greatest interest for simple and rapid chemical or biological agent detection since detecting color changes in applications such as wearable sensing, despite fabrication challenges.²⁹ Such colorimetric sensors could function with a simple LED as an optical source, further emphasizing the simplicity and utility of PhC sensors. For ease of reference, Table 1 summarizes bulk and surface sensitivities of various PIC-based sensors and the representative analyte detected.

Table 1

Comparison of PIC-based sensors for biochemical detection. Adapted from Gavela et al.10

While many varieties of PIC sensor designs being explored, it is unlikely that there will be a single design globally favored. Instead, application-specific single-ID PIC sensors will be developed to ensure that specific DoD mission-needs are satisfied.

5.1.

Fully Monolithic Photonic Chips

Although significant advancements have been made in the CMOS compatible fabrication of PIC technologies, fully integrated monolithic silicon photonic/electronic chips with on-chip sources, detectors, and amplifiers have yet to be demonstrated. This is primarily a material limitation because silicon is an indirect bandgap semiconductor with extremely low light emission efficiency. Given the vast knowledge and infrastructure for silicon device fabrication and the extensive role of silicon in electronics, it is desirable to find ways to use a silicon platform for all-optical devices.

To that end, several efforts have been made to enhance silicon emission efficiency to demonstrate on-chip lasers, such as growing successive layers of indium arsenide on silicon to form quantum dots as demonstrated through the DARPA Electronic-Photonic Heterogeneous Integration (E-PHI) program.^44,45 This demonstration is now serving as a foundation for the development of other photonic components such as optical amplifiers, modulators, and detectors.

Current efforts are now focused at large-scale integration of electronics and photonics, such as the recent DARPA Photonically Optimized Embedded Microprocessors (POEM) program.⁴⁶ This program is developing chip-scale integrated photonic technology to enable seamless intrachip and off-chip photonic communications. This program has already had several groundbreaking demonstrations, including the first single-chip microprocessor with 70 million transistors and 850 integrated photonic elements, all fabricated using the standard microelectronics foundry process, to work together to provide logic, memory, and interconnect functions.⁴⁷ This is significant because the silicon fabrication requirements for electronics and photonics are orthogonal—while electronics requires very thin silicon materials (order of single nm) to quickly dope and transition the silicon between conducting and insulating states, photonics requires much thicker silicon (order of 100 nm) to confine light efficiently. Earlier this year, the same group demonstrated, for the first time, integrated photonics (optical waveguides and resonators, transceivers, high-speed optical modulators, and sensitive avalanche photodetectors) with silicon nanoelectronics in polycrystalline silicon operating at $10\mathrm{Gbits}/\mathrm{s}$, all on the same chip.⁴⁸ These developments are pushing the technological limits of integrated photonics, but they do not yet include on-chip optical sources.

While the on-chip integration of sources, detectors, and electronics, may be possible in the near-IR, there is still much interest in integrating different material platforms, which may also enable devices operating at longer wavelengths, such as the mid- and long-wave IR. For example, photonic systems based on III–V materials operating in the mid-IR such as GaAs, InP, and GaN already have a complete set of photonic components, including on-chip monochromatic lasers, amplifiers, and modulators.⁴⁹ These platforms are continually being developed to create hybrid platforms with silicon photonics.

5.2.

High-Volume Manufacturing of PIC Devices

The prospect of single-chip, multitechnology integration opens an array of opportunities. While on-chip PIC components are still under development and being optimized using several material platforms, the National Network for Manufacturing Innovation has founded the American Institute for Manufacturing Integrated Photonics (AIM Photonics) to advance large scale manufacturing of CMOS-compatible PICs. This institute serves to leverage existing CMOS electronics infrastructure and fabrication technologies to develop a complete PIC ecosystem within the United States; this includes foundry access, automated design tools, resources for packaging, assembling and testing devices, and a domestic workforce.⁵⁰

AIM Photonics is currently focusing on achieving a manufacturing readiness level 7, as defined by the DoD Manufacturing Technology program, for CMOS-compatible integrated electronic and photonic devices, such as the device shown in Fig. 6.⁵¹ CMOS devices based on both silicon and indium phosphide would span applications in telecom, RF analog circuits, and PIC sensors and array technologies. Integrated devices should serve to increase both the capacity and reliability while decreasing the cost, size, weight, and power (SWaP) consumption of devices.⁵² AIM also receives funding from the Navy, to advance their multiproject wafer assembly and packaging service, which provides access to the most advanced 300-nm semiconductor fabrication facilities and a suite of passive and active PIC wafer processing technologies. This infrastructure, largely funded by the Government, will ensure that the necessary manufacturing resources are available for mission-critical technologies.

Fig. 6

PIC sensor device with integrated photonics and electronics manufactured by the AIM Photonics teams.

AIM Photonics is already making an impact on mission-relevant technologies, including the successful development of a chip-scale Sarin gas photonic sensor.⁵³ The team achieved the necessary high sensitivity by developing a long-path on-chip waveguide and coating it with a polymer to concentrate the simulated Sarin agent. Their next steps are to develop a fully integrated sensor with on-board spectrometers and lasers and to apply such monolithic sensors for nondefense applications.

In collaboration with the MIT Microphotonics Center and the International Electronics Manufacturing Initiative, AIM released their 2017 Integrated Photonics Systems Road Map in March 2018.⁵⁴ The roadmap, which serves to drive technology development, identifies several grand challenges, including: the drastic improvement in power efficiency, the development of tools for large-scale PIC system design, new manufacturing and material technologies, and standards for interoperability to broaden market potential. AIM has already solicited and funded several proposals for R&D efforts to overcome these challenges, including a new effort to develop a CMOS-compatible waveguide platform for integrating mid-IR and long-wave IR on-chip laser sources (spanning 3 to $14\mu \mathrm{m}$). The large wavelength range would enable several DoD-relevant applications including detecting atmospheric trace gases. We discuss the importance of these integrated sources further below.

5.3.

Broadband, On-Chip IR Sources for Biochemical Detection

Looking specifically toward spectroscopy-based biochemical sensing platforms, there is a clear need for broadband sources with extremely high-spectral resolution, different from the on-chip monochromatic lasers mentioned above: increasing the bandwidth will allow for more properties of each analyte to be measured, resulting in improved specificity and multiplexing. One promising technology is optical frequency combs, which have emerged in the past decade as powerful and precise tools with coherent, equidistant spectral lines spanning a broad range with unprecedented precision in frequency in timing.⁵⁵ Several programs across the DoD have been pushing the development of frequency combs spanning the visible and IR ranges for applications in biochemical sensing, time-frequency transfer, and metrology, such as DARPA Spectral Combs from UV to THZ (SCOUT),⁵⁶ DARPA program in Ultrafast Laser Science and Engineering (PULSE),⁵⁷ and the DARPA Direct On-Chip Digital Optical Synthesizer (DODOS)⁵⁸ programs. These programs continue to push the limits of frequency combs in terms of power generation, bandwidth, spectral resolution, and detection.

One major challenge of optical frequency combs is their requirement for large spectrometers to spectrally resolve each individual comb tooth. However, by interfering two combs with slightly different repetition rates, one can generate an RF comb composed of heterodyne beat pairs, which contain all the relevant spectral information and is easily accessible with RF electronics.⁵⁹ This technique, termed dual-comb spectroscopy (DCS), has several advantages over conventional spectrometers, including a broad spectral coverage, high-frequency resolution, high-sensitivity/signal-to-noise ratio, and fast acquisition speed; as additionally, DCS performance is not limited by the instrument OPL, which may enable ultracompact systems. Hence, DCS is proving to be an invaluable tool for biochemical detection in the near- and mid-IR.

There is also a strong push to move sources for chemical detection to longer wavelengths in the mid-IR and long-wave IR ranges as it allows for probing much stronger fundamental molecular vibrations, as opposed to the overtones excited in the near-IR. Programs such as SCOUT have specifically pursued the development of mid-IR frequency combs for DCS-based biochemical sensing and have demonstrated significant enhancement in sensitivity and limits of detection. Earlier this year, Ycas et al.⁴⁰ at the National Institute of Standards and Technology demonstrated a mid-IR dual-comb system spanning 2.6 to $5.2\mu \mathrm{m}$ with sub-MHz frequency precision. While this is an open-path-based system, recently developed chip-scale battery-operated frequency combs have the potential to revolutionize the SWaP of DCS systems.⁶⁰ Just this year, Gaeta et al. have demonstrated high-SNR absorption spectroscopy with an on-chip DCS for the first time in the NIR, with a 170-nm bandwidth and a $20\text{-}\mu \mathrm{s}$ acquisition time. Further broadening of the spectral coverage of this on-chip system and shifting to longer wavelengths could enable truly disruptive chemical and biological sensing technologies relevant to DoD applications. However, moving to the mid-IR does introduce several operational challenges including limited achievable $Q$-factors and increased water vapor absorption, which, respectively, reduce sensitivity and increase background signatures, making the detection and quantification of analytes in native environments more difficult. We discuss the challenges with detection in native environments in further detail below.

5.4.

Analyte Identification in Native Environments

As PIC-based sensors transition from the lab to native environments for field use, metrics for assessing utility must go beyond sensitivity alone to also include concepts of device robustness and trustworthiness, such as the probability of correct detection, false alarm rate (FAR), and the response time.⁶¹ FAR is defined as the number of false alarms per unit time and defines how well the sensor identifies analytes amidst a background and interferents. For biochemical detection, false negatives are far more dangerous than false positives, though a high false positive rate undermines the detection system trustworthiness (a challenge to combat operational and response fatigue). The sensor’s receiver operating characteristic (ROC) curve captures the performance trade-off between sensitivity, probability of correct detection, and FAR and dictates an implicit associated detection confidence.⁶¹ Additionally, metrics for long-term stability of devices, such as the Allan variance or Allan deviation, will be required to understand device stability to noise processes and drift and assess their ability to transition to DoD-relevant missions. We direct the reader to Ref. 61, a government-issued study with in-depth analyses of chemical and biological sensor metrics, testing, and recommendations.

The DoD is actively using machine learning algorithms that are trained with experimental data to reduce FAR by identifying analyte signatures amidst a strong background and interferents; this includes water absorption in the mid-IR, which obscures real-time analyte identification. In terms of fielded devices, DHS S&T first responders group is actively looking at machine learning and artificial intelligence technologies for wearable sensor technologies.⁶² Their new platform, assistant for understanding data through reasoning, extraction, and synthesis, collects data from various sensors and notifies first responders of any necessary actions to be taken, as well as any contextual insight of similar events that occurred elsewhere in the country.⁶³

In addition to machine learning algorithms, networking orthogonal sensing platforms to reduce the overall FAR across a network are a capability being developed at DARPA through the SIGMA+ program.⁶⁴ SIGMA+ is focused on both developing highly sensitive detectors and advanced intelligence analytics to detect minute traces of substances related to CBRNE threats. By fusing CBRNE data, as well as local weather and external contextual data across a network infrastructure, SIGMA+ aims to develop a persistent, real-time early detection system to monitor a city-wide region (${\mathrm{km}}^2$) with maximum sensitivity and minimum FAR.