1. Introduction

The ongoing conversion of indoor lighting to energy efficient LED systems offers enormous opportunity for increasing the functionality of lighting from today's modest on/off/dimming control to a new Smart Lighting paradigm that takes advantage of LEDs' electronic compatibility and flexibility. This new lighting paradigm includes lighting for enhanced worker/student productivity, health effects such as circadian entrainment reinforcing the human sleep/wake cycle, visible light communications (VLC) to alleviate the growing wireless bottleneck, and occupancy/activity sensing to provide “the right light where and when you need it [1].”

The highest lighting efficacy will be achieved with multiple LEDs at different colors across the visible, eliminating the energy losses inherent in phosphor color conversion. VLC will require a multiple-input multiple-output (MIMO) architecture with multiple LEDs from multiple fixtures to provide the necessary Gbps data rates and to support mobility as people move with their personal devices [2]. Light has many impacts on health and productivity; spectral as well as intensity variations are important for optimizing the human environment [3,4]. An even greater energy savings, along with a more comfortable experience, is available by adapting lighting to human activity in addition to the savings from the improved efficacy of LEDs.

All of these impacts suggest that the end point of the ongoing LED conversion will be arrangements with multiple LEDs along with occupancy and activity sensing. A plenoptic sensor technology matched to this source paradigm will require both spectral (30 nm) and angular (~100 mrad or 5°) resolution. The technology must necessarily be based on a scalable silicon IC platform to meet mass-market cost targets; devices will require a low profile for broad installation flexibility, and cannot have any moving parts such as a grating rotation. For the VLC application, > 20 dB out of band rejection in both angle and spectral dimensions is required to avoid inter-channel, background and multi-path interference effects [5,6].

Today's digital cameras are designed to mimic the human eye response, which evolved to match a continuous solar spectrum, with three spectrally overlapping stimulus channels [7]. Colors are determined by the tri-stimulus intensity ratios. Commercially available dye-based color sensors [8] mimic this spectral response with broad and overlapping spectral channels. Color sensors based on interference filters have limited angular range and employ a non-scalable manufacturing technology [9]. Neither of these technologies meets the resolution and crosstalk requirements inherent in a spectrally discontinuous, information-broadcasting, multi-LED, multi-fixture source array.

There have been many reports aimed at using surface plasma wave (SPW) filters as a replacement for the color dyes in digital cameras, motivated by the reduced manufacturing costs of a single lithography step for the three color pixels and the trend to smaller pixels where the required dye thickness becomes more difficult to achieve [10–14]. Generally, limitations of the SPW approach include: 1) the relatively high metal optical losses in the visible restrict the available bandwidths; spectral widths are typically 100- to 200-nm, an order of magnitude larger than the desired bandwidths for smart lighting; and 2) the transmission is low, typically no larger than 10%, limiting the sensitivity of the measurement.

Guided-mode resonance (GMR) filters, consisting of a grating coupler and a single mode waveguide slab have demonstrated both angular and spectral sensitivity in reflection and transmission [15,16]. Off-resonance, GMR filters simply act as a dielectric medium, usually with the majority of the incident power simply being transmitted. On resonance, the grating couples some of incident photons into the waveguide and the propagating photons in the waveguide are coupled back into the reflected and transmitted beams. As a result of the phase shifts inherent in this process, the re-radiated photons reinforce the reflected wave and interfere destructively with the directly transmitted light to reduce the transmitted power. Since the waveguide is lossless and the grating is large (many wavelengths), an extremely narrow resonance response is achieved.

Waveguide integrated optics at telecommunications wavelengths has demonstrated that grating coupling into waveguide modes can provide the necessary spectral and angular filtering [17–19] with recent demonstrations of only 0.6 dB loss in conversion from a 2D waveguide to a single mode fiber. We adopt this concept to the visible, using lossless, CMOS-compatible SiO₂/Si₃N₄/SiO₂ waveguides integrated with conventional Si photodetectors. This technology easily meets the spectral and angular resolution requirements, is readily scalable to array architectures, and easily will provide the RF bandwidths and out-of-band rejection required for VLC.

2. Waveguide detector element

The concept of integrated waveguide spectral detection is shown schematically in Fig. 1(a). The incident light is coupled to the waveguide at a specific wavelength and incident angle, providing a spectral/angular filtering function, and out-coupled from the waveguide downstream from the coupling region to a photodetector fabricated in the underlying silicon substrate. This structure is related to the GMR filter with a detector integrated onto the same platform as the waveguide. In this device, the gratings are of finite size, comparable to the coupling/re-radiation length, and the power coupled into the waveguide is detected as opposed to the reflection/transmission of the GMR filter. As noted below, the resolution is a function of the width of the illuminated grating along the coupling direction as well as the grating coupling strength and can be adapted to fit the resolution requirements of the application. A plenoptic sensor will require an array of such detection elements with different gratings and orientations. A visual demonstration of the propagation is shown in Fig. 1(b) where a 532-nm laser source is used. The laser is incident on the waveguide at a position offset from the photodetector element. In this experiment, there was a grating over the entire area between the incident laser spot and the detector element, the out-coupling from the waveguide is evident as the line of scattered light leading from the laser spot to the detector element.

 

Fig. 1 (a) Schematic of the angle-of-incidence and wavelength integrated sensor concept; (b) Demonstration of waveguide coupling and propagation for a green laser source. The grating extends across the entire device region and the light coupled from the waveguide mode to free space is observed. The cover over the detector area to eliminate direct illumination of the detector element is not yet implemented. See text for details of the waveguide and grating structure.

Download Full Size | PDF

The grating coupler along with the single mode slab waveguide provides the necessary angular/wavelength selectivity as is evident from the coupling equation:

A SiO₂ (n_SiO2~1.5) lower cladding with a thickness of 1 μm was deposited to eliminate leakage into the silicon substrate assuring low waveguide losses. The Si₃N₄ guiding layer was ~200 nm thick (n_Si3N4 ~1.8) and the top cladding was adjusted to control the coupling strength. For the measurements reported here the top cladding was SiO₂ with a thickness of ~30 nm. In separate devices, 320- and 380-nm gratings were fabricated by interferometric lithography over large areas of the wafer [20]. Two probes were used to make electrical contact to the device contact pads. The optical components (laser, lens, apertures, polarizer) were mounted on a optical rail attached to a computer controlled rotation stage for the angular measurements as shown in Fig. 2.

 

Fig. 2 Experimental arrangement. All of the optical components are mounted on a computer controlled arm that rotates about the top grating on the waveguide at a position offset from the photodetector junction. Contacts are etched and metalized on the top surface of the silicon wafer and probes are used in these initial experiments.

Download Full Size | PDF

The photodiode is an avalanche PN detector fabricated on a silicon wafer using a standard CMOS fabrication process. The waveguide structure over the PN detector is also fabricated using CMOS compatible processes. The doping concentration of the p⁺ substrate was 5x10¹⁷ cm⁻³. The n⁺ detector region was fabricated with a diffusion process with doping concentration of 1x10¹⁹ cm⁻³. The p and n regions are highly doped in order to reduce the breakdown voltage so that an avalanche multiplication gain can be obtained at lower voltage, compatible to commercial CMOS circuit applications. The breakdown voltage of the CMOS avalanche photodiode (APD) is 8.5V with the maximum gain, M, of about 75. The junction depth is 0.5 μm and the detector area is 200 × 200 μm². Details of the fabricated CMOS compatible APD have been presented elsewhere [21]. To maximize the accuracy of the photocurrent measurements in the experiments presented here, the photodetectors were unbiased so there was no gain. This eliminates uncertainty due to avalanche gain variations and bias dependencies in APDs. The overall system sensitivity depends on the spectral variation of the ambient light as well as the coupling efficiency into and out of the waveguide and the details of the Si p-n junction photodetector; the option of avalanche gain extends the sensitivity to photon counting levels if necessary.

3. Experimental results

For initial testing of the waveguide filtered CMOS compatible photodetector, diode-based, multi-mode RGB lasers of wavelengths 652.3-, 532.2- and 407.8-nm were used. The experimental setup consists of the laser light source followed by an infrared filter, polarizer, long focal length lens and an aperture to provide uniform illumination across the ~200 × 200 μm² coupling region and avoid any direct illumination of the junction region (the cover protecting the detector element from direct illumination shown on Fig. 1(a) has not yet been implemented). Incoming light at the resonant wavelength and angle is scattered by the grating and couples into the waveguide, propagates to the junction area, and is decoupled into the photodetector. Out-of-resonance light does not couple into the waveguide and is either reflected or transmitted into and absorbed in the silicon far from the photodetector active area and does not contribute to the photocurrent. The illumination angle of the incident beam relative to the grating was scanned with a resolution of 6 arcsec. Figure 3 shows the ratio of the measured photocurrent to the incident power of each laser source, normalized in each case to the peak measured photocurrent, for two sets of measurements with the same waveguide structure but different grating pitches of 320- and 380-nm. The results for the two pitches are vertically offset for clarity. The measured angular linewidths [< 0.5° corresponding to a wavelength spread of ~3 nm ($\Delta \lambda \approx -d\cos \theta \Delta \theta$) from Eq. (1)] are slightly wider than the theoretical predictions (discussed below) and show some fine structure, probably corresponding to the multi-mode character of the lasers. The linewidth is a convolution of the scattering/absorption propagation losses of the bare (no grating) waveguide and the spectral width corresponding to the length of the coupling area (the smallest of the width of the grating, the illumination spot size, or the coupling length). In these experiments, the laser spot size(~200 μm) along with the intrinsic bandwidth of the multimode laser sources are the dominant contributions to the observed linewidth.

 

Fig. 3 Angular resolution of a grating coupled waveguide detector for RGB laser sources. The expanded views show the TE (solid) and TM (dotted) experiments and simulations (black lines).The bottom panel shows the results with the same waveguide with grating periods of 320- and 380-nm. The 320-nm grating results are offset and expanded in the top panels and compared with simulation.

Download Full Size | PDF

4. Theory and simulation

Figure 4(a) shows the wavelength and angle for coupling for a range of grating periods. For the longer period gratings, multiple orders are included as indicated (e.g. 800/2 refers to the second-order of a 800 nm pitch grating, etc.). The measured points from Fig. 3 as well as an observed second-order peak for the 380-nm grating in the blue are indicated. Figure 4(b) shows the required grating periods for a fixed angle of incidence. Importantly, all of the required grating periods are easily within the range accessible to modern lithography tools so that full arrays of devices can be readily manufactured with a single additional lithography step once the waveguide layer structure has been deposited atop a CMOS-compatible detector array.

 

Fig. 4 (a) Wavelength vs. coupling angle for different grating periods. Multiple orders of the grating are shown. (e.g. the notation 800/2 refers the second order of a 800 nm pitch grating). The experimental points are indicated; (b) Wavelength vs. grating period at a fixed angle (periods are indicated at the top of the figure, the fixed angles are indicated at the bottom of the figure. Both forward and backward scattering regimes are indicated. For both figures the solid lines are TE modes and the dotted lines are the TM modes.

Download Full Size | PDF

The coupling strength is an important parameter since the goal is to maximize the power coupled into the waveguide. As indicated by the measured radiative decay lengths, our present structures are undercoupled and larger signals would be available with an improved coupling geometry. In the coupled-mode approximation (considering only a single waveguide mode and the incident and transmitted/reflected fields) the intensity of the modal output relative to the input plane wave intensity for grating coupling is given by [22]:

Figure 5 shows a numerical evaluation of the intensity of the mode coupled into the waveguide (relative to the intensity incident on the coupling region) for a grating etched into the SiO₂ top cladding for two coupling widths, assuming a broadband source that overfills the coupling region. A finite-difference time-domain (FDTD) simulation (Lumerical FDTD software) was used to optimize the structure. Plane wave excitation was assumed, and perfectly matched layer boundary conditions at the edges of the simulation area were used to avoid reflections. A plane wave containing a broadband spectrum across the visible (400- to 700-nm) is taken as the source. For this calculation, the cladding thickness is taken as 150 nm, and the etch depth from the top of the cladding is varied from 50 to 150 nm in 10 nm steps with increased coupling corresponding to a deeper etch. Several noteworthy trends are apparent: 1) for weak coupling, where re-radiation back into free space can be ignored, the linewidth scales inversely with the coupling region width as expected; 2) in this weak coupling regime, the modal intensity increases quadratically with the width of the coupling region (the lowest, rightmost curves in each figure); 3) there is a maximum in the modal intensity as the coupling strength is increased, this is a result of the re-radiation of the waveguide light into the reflected/transmitted fields leading to a saturation of the power in the waveguide; 4) the linewidth increases as the re-radiation becomes more significant and 5) the shift in the resonance wavelength is a result of the decrease in the cladding effective refractive index for the deeper gratings providing the stronger coupling, which shifts the modal index.

 

Fig. 5 Simulated intensity in the waveguide relative to the incident intensity for two different coupling widths. The depth of the grating into the top cladding is the parameter with a deeper etch corresponding to an increased coupling, see text for details. The drop in the peak intensity at high coupling strength is due to re-radiation into the transmitted/reflected beams.

Download Full Size | PDF

5. Future directions

There are many directions for further optimization of these results. The use of chirped or multiple shorter gratings will allow adjustment of the spectral/angular width to adjust the spectral width to match LED source linewidths of 20- to 30-nm and to assure complete spectral coverage with a relatively small number of parallel channels (with different grating pitches). As a result of the re-radiation of the waveguide power, the coupling strength has to be tuned to the total grating width. The output coupler can have a higher coupling strength than the input coupler to concentrate the waveguide power into a smaller detector area and allow a smaller, faster photodetector. Curved gratings will further focus the coupling energy allowing further decreases in the detector area, effectively creating a 2D lens.

An array of similar detectors with different period and orientation gratings, all well within current lithographic tool capabilities, will provide the needed plenoptic spectral and angular coverage. One strategy, suggested by Fig. 3(b), is to arrange for only a relatively narrow range of incident angles, for example around normal incidence and build an array of grating-coupled photodetectors to directly monitor the spectrum. From Fig. 3(b) a grating period range of 420- to 215-nm will be sufficient to cover the full visible spectrum. Alternatively, one coupling area can provide inputs for several detectors at different wavelengths. Assuming a 20 nm resolution this requires just 15 distinct (different grating period) waveguide-coupled detectors. For a simple grating two different wavelengths can be directed in opposite directions as shown in Fig. 6. Adding a second grating in an orthogonal direction allows simultaneous monitoring of four different wavelengths; clearly this multiplexing can be continued with more complex grating structures to provide a complete spectral readout for a fixed angle at any desired resolution set by the widths of the coupling area. This provides an exciting new platform for many spectroscopic applications.

 

Fig. 6 (a) Dual detector configuration with a single input grating area. (b) separation of red (652.3 nm; backward coupled) and green (532.2 nm; forward coupled) wavelengths incident on the same spot of the grating coupler.

Download Full Size | PDF

In other scenarios, it will be important to sample the spectrum as a function of angle. The simplest solution is use apertures to choose a fixed polar angle, θ, say 30° and then built a separate array for a number of azimuth angles (e.g. every 45° for φ = 0° to 360°) for a total of eight individual angular arrays. In this scenario, the required grating pitch varies from 330- to 170-nm for backward coupling and from 600- to 325-nm for forward coupling. As above, the input grating section can be complex to separate many wavelengths into different detectors, providing a full plenoptic analysis.

A second architecture, needed for the VLC application, would provide optical access across a significant solid angle so that different wavelengths would couple into each waveguide at different angles. As a result there will be a spectrum of light at each detector. There are both electronic and physical optics approaches to sorting out the spectral information. Since we are dealing with “smart” lights that will be broadcasting information, they will necessarily be broadcasting identification signals than can be used to distinguish the different wavelengths. In a simplistic scenario, each set of diodes at a specific wavelength in a fixture will be modulated at a different frequency, too high for human perception, but easily sorted out electronically on chip. Alternatively, a physical optics solution is to provide a wavelength dispersive element along the waveguide with multiple detectors. One approach is to adiabatically taper the waveguide to a single transverse mode guide and then use either an array waveguide grating, or a series of ring resonators, to provide the spectral resolution. This will also allow sensing of natural (sun) light illumination and allow the lighting system to adapt to variations in background illuminance. For this general situation requiring full angular and spectral information, certainly the case for the VLC application where the sensor might be located in a mobile device, will require an array of ~15 × 9 × 36 (20 nm in wavelength × 5° in θ × 10° in φ) = 4860 elements, modest in a world of inexpensive 20 Mbit consumer cameras. Geometries with 2D gratings and individual detectors arrayed in a circle around the collection area will reduce the number of coupling gratings and the overall area of the sensor.

6. Summary and conclusions

LEDs offer a vastly richer set of functionalities than traditional incandescent and fluorescent light bulbs. Harnessing this functionality, which impacts human health, well-being, and productivity as well as data communications, will require a Smart Lighting paradigm with an extensive array of sensors and feedback in place of today's simplistic, open-loop lighting control. One specific requirement is a plenoptic (angle and spectral sensor) to support a wide color gamut and the communications requirement for a MIMO architecture to provide mobility and aggregate bandwidth. Today's color cameras have only three overlapping color channels and do not provide the required angular/spectral resolution and out-of-band signal rejection. We have demonstrated a new detector concept, a grating coupled waveguide with a silicon photodetector offset from the input coupling region that provides the necessary angular/spectral resolution and meets the requirements for smart lighting. An angular (spectral) resolution of ~1° (5 nm) has been demonstrated. Incorporation of these detector elements into an array to demonstrate spectral measurements of LEDs is underway. This waveguide coupled, CMOS compatible, detector construct opens a rich assortment of architecture possibilities that we have only begun to explore.

There are many other applications for inexpensive, compact, low profile, robust, vibration-insensitive, integrated, solid-state visible spectrometers ranging from chemical and environmental monitoring to industrial process control. The plenoptic detector array we have described will find application across many of these areas. The spectral resolution can be structurally tuned across a wide range from less than 1nm to more than 100 nm providing an wide range of functionality. The use of avalanche photodetectors can extend the sensitivities to photon counting applications. The same concept can be extended to other wavelength ranges with different detector material systems, extending the range of applications.

The Si fabrication process is fully CMOS compatible, the junction and the waveguide structure are all standard processes using CMOS compatible materials. The lithography demands of the gratings are well within current industry capabilities. The sensing electronics will be fabricated alongside the detector array, providing a fully integrated, and readily manufacturable, solution for LED lighting and other applications.