Key Research Milestones
i) Nanophotonics
ii) Photonic Information Processing
 Wordrecord low VpL = 0.01Vmm MZI modulator in ITO plasmon Si photonics (2018)
 First 2D material integrated modulator microring resonator modulator and photodetector (2018)
 First ITObased MZI modulator with VpL = 0.5Vmm (2018)
 Fastest (60 GHz) Graphene modulator (2018)
 Demonstrated Siliconbased integrated tunnel source operating at room temperature (2018)
 Demonstrated 100 aJ/bit Graphene Plasmon Modulator in Si Photonics (2017)
 Developed electricallydriven siliconbased plasmon source operating at room temperature (2017)
 Designed 2x2 <10um compact Electrooptic Switch (2015), and NxN crossbar router (2017)
 Demonstrated ENZ ITO modulator (2012) and unity index tunability in ITO films (2015)
 Proposed (2008) and demonstrated (2011) hybridphotonplasmon mode
 First semiconductor plasmon laser (2009) licensed for $2M to major electronics manufacturer
ii) Photonic Information Processing
 Developed alloptical feedforward neural network (2018)
 Developed photonic convolutional neural network accelerator (2018)
 Developed highspeed photonic digitaltoanalog converter (DAC) (2018)
 Developed a photonicaccelerated 4f Joint Transform Correlator for convolution processing (2018)
 Demonstrated Reconfigurable Photonic Processor (2018)
 Proposed electrooptic modulators as nonlinear activation function in neural networks (2017)
 Demonstrated identifying mirror symmetry with delay in spiking neural networks (2017)
 First photonic integrated temporal FFT chip, with demonstration currently ongoing (2017)
 Developed a photonic reconfigurable graph processor (2017)
 Developed and analyzed cognitive and DDDASbased hybrid photonic NoC (2017)
 Developed electronic photonic, plasmonic hybrid NetworkonChip (2016)
Hybrid Integrated Photonics (HIP)
Hybrid Integrated PhotonicsA fundamental challenge for optics and photonics are the weak interaction between light and matter. This can be enhanced in 2 ways; (i) increasing the interaction time deploying resonators, and (b) providing a lengthscale match between the optical wavelength and the wavefunction of electrons in matter. We follow a combination of both which leads to strong optical confinement factors quantified by the ratio of the cavity Qfactor divided by the optical mode volume – known as the Purcell factor, Fp. Our vision is to demonstrate attojouleperbit efficient optoelectronic devices with highspeed modulation or reconfiguration potential. The latter has applications not only for optical data communication, but more interestingly for optical algorithms such as neuromorphic compute engines.

2D material Printer:Towards Large scale IntegrationAs we enter into the era of functional scaling where the crossroads of MoreMoore and MoreThanMoore meet, the search for new functional devices enabling novel materials become very critical. 2D materials could be promising platform since they exhibit exciting properties as compared to their bulk counterpart. Over the last decade, intense research has been mainly focused on physics and application of these novel 2D materials. However, one main challenge prohibiting such materials from large scale technology implementation is the flake to wafer integration keeping the quality of the material intact. Hence, Precision and chip contaminationfree rapid placement of twodimensional (2D) materials is expected to overcome the bottleneck of large scale application.
Here, we demonstrate a novel method of transferring 2D materials resembling the functionality known from printing; utilizing a combination of a sharp microstamper and viscoelastic polymer, we show precise placement of individual 2D materials resulting in vanishing crosscontamination to the substrate. Our 2D printermethod results in an aerial crosscontamination improvement of two to three orders of magnitude relative to stateoftheart transfer methods from a source of average area for single flake (~50 µm^2). Moreover, we find that the 2D material quality is preserved in this transfer method. Testing this 2D material printer on tapedout integrated Silicon photonic chips, we find that the microstamper stamping transfer does not physically harm the underneath Silicon nanophotonic structures such as waveguides or microring resonators receiving the 2D material. We further demonstrate functional devices such as Graphene tunnel junctions and transistors, and TMDbased material tunable microring resonators. Such accurate and substratebenign transfer method for 2D materials could be industrialized for rapid device prototyping due to its high timereduction, accuracy, and contaminationfree process. 
Heterogeneous Integration of 2D materials on Si Photonic PlatformHeterogeneous integration of ultrathin atomic layer is expected to dominate future photonic and electronic devices due to their rich unique physical properties for advances in optical communication technology. Hence, a detailed understanding of the loss and index change on Si integrated platform is a prerequisite for advances in optoelectronic devices impacting optical communication technology, signal processing, and possibly photonicbased computing. Here, we present an experimental guide to characterize transition metal dichalcogenides (TMDs), once monolithically integrated into the Silicon photonic platform at 1.55 mm wavelength. We describe the passive tunable coupling effect of the resonator in terms of loss induced as a function of 2D material layer coverage length and thickness. We further demonstrate a semiempirical method to determine the index of a 2D material (nMoTe2 of 4.36+0.011i) near telecommunicationrelevant wavelength by integrating few layers of MoTe2 onto a microring resonator. We also find the TMDCstoring’s circumference coverage length ratio required to precisely place the ring into critical coupling condition to be about 10% as determined from the variation of spectral resonance visibility and loss as a function of TMDCs coverage. The placement, control, and opticalproperty understanding of 2D materials with integrated photonics paves a way for further studies of active 2D material based optoelectronics and circuits.

Fundamental Scaling Law of NanophotonicsThe success of information technology has clearly demonstrated that miniaturization often leads to unprecedented performance, and unanticipated applications. This hypothesis of “smallerisbetter” has motivated optical engineers to build various nanophotonic devices, although an understanding leading to fundamental scaling behavior for this new class of devices is missing. Here we analyze scaling laws for optoelectronic devices operating at micro and nanometer lengthscale. We show that optoelectronic device performance scales nonmonotonically with device length due to the various device tradeoffs, and analyze how both optical and electrical constrains influence device power consumption and operating speed. Specifically, we investigate the direct influence of scaling on the performance of four classes of photonic devices, namely laser sources, electrooptic modulators, photodetectors, and alloptical switches based on three types of optical resonators; microring, FabryPerot cavity, and plasmonic metal nanoparticle. Results show that while microrings and FabryPerot cavities can outperform plasmonic cavities at larger lengthscales, they stop working when the device length drops below 100 nanometers, due to insufficient functionality such as feedback (laser), indexmodulation (modulator), absorption (detector) or field density (optical switch). Our results provide a detailed understanding of the limits of nanophotonics, towards establishing an optoelectronics roadmap, akin to the International Technology Roadmap for Semiconductors.

Scaling Vectors of Attojoule per Bit Modulators for Hybrid Plasmon Graphene Modulator on Silicon PhotonicWe demonstrate the first hybrid silicon photonics plasmonic electroabsorption using Graphene. The unitystrong index change in Graphene based on the Pauliblocking mechanism outweighs the small optical overlap factor. The plasmonic hybridsilicon mode allows optical concentration and high group index which enhance modulation efficiency. We demonstrate sub 1Volt efficient modulator of 8 um in length with a energyperbit function <200 aJ/bit.

Graphene Slot Plasmonic EAM DesignWe demonstrate a plasmonic slot waveguidebased Graphene modulator on a Silicon photonics platform. Here the extinction ratio can exceed 1dB/um since the field polarization of the slot is inplane with Graphene. Our next steps are to create the Graphenebiasing capacitor in a pushpull configuration both above and below the slot. This enables about 4dB/um strong modulation and allows for recordcompact devices lengths (750nm = λ/2). We will integrate this slot into an SOI waveguide platform next.

OnDemand Single Photon Sources based on Quantum Tunneling on SiliconAn efficient siliconbased light source presents an unreached goal in the field of photonics, due to Silicon’s indirect electronic band structure preventing direct carrier recombination and subsequent photon emission. Here we utilize inelastically tunneling electrons to demonstrate an electricallydriven light emitting siliconbased tunnel junction operating at room temperature. We show that such a junction is a source for plasmons driven by the electrical tunnel current. We find that the emission spectrum is not given by the quantum condition where the emission frequency would be proportional to the applied voltage, but the spectrum is determined by the spectral overlap between the energydependent tunnel current and the modal dispersion of the plasmon. Distinct from LEDs, the temporal response of this tunnel source is not governed by nanosecond carrier lifetimes known to semiconductors, but rather by the tunnel event itself and Heisenberg’s uncertainty principle. We are now investigating intensions of this light source utilizing the Coulombblockade to generate single electron tunneling events towards demonstrating ondemand single photons on this Silicon platform.

ITORing CouplingModulator on Silicon PhotonicsMicro ring resonators are extensively used in in integrated photonic applications and are highly sensitive to electrooptical effects. Majority of available ring based modulators are interactivity modulation where active region changes the phase or absorption of the stored optical mode in the cavity. Such devices are essentially limited by the photon lifetime in the cavity. In contrast, coupling modulation devices can change the crosscoupling coefficient of the resonator and take advantage of the nonquasistatic modulation regime. We demonstrate an electricallydriven, CMOS compatible, ring resonator coupling modulation mechanism based on tuning of free carriers in Indium Tin Oxide (ITO). The modulator device consists of two ITO layers separated by oxide fabricated on the coupling region of silicon ring resonator. We are investigating modulation performance of such CMOS compatible coupling modulation devices.

Optical Network on Chip (ONoC)Continuing demands for increased compute efficiency and communication bandwidth have led to the development of novel interconnect technologies with the potential to outperform conventional electrical interconnects. We benchmark various interconnect technologies including electrical, photonic, and plasmonic options. We contrast them with Hybrid Photonic Plasmonic Interconnect(s) (HyPPI) where we consider plasmonics for active manipulation devices, and passive photonic waveguides for signal propagation. Our result shows that for a pointtopoint link, HyPPI provides over 10x to 100x performance improvement in bandwidth, delay, energy efficiency and packing density on chip, which makes HyPPI a great candidate for edge computing for Internet of Things (IoT).
Another application for HyPPI is the Dynamic Data Driven Application System (DDDAS), where computations and measurements form a dyanamic closed feedback loop in which response to the changes in the environment. Our HyPPI based Dynamic Data Driven Network onChip (D3NoC) shows up to 89% latency and 67% dynamic power improvement over traditional electrical networks in a 16x16 mesh network size. 
Optical Algorithms & Information Processing (OAIP)
Photonic AlgorithmsIn order to understand the benefits and tradeoffs of optics for computing and signal processing, we divide he compute space until 3 orthogonal axis. (1) Computational complexity defines the ability to access a memory; (2) Information representation distinguishes between a finite symbol set, continuous states, or metaphorics; (3) Reconfigurability relates to the machines learning capability. Since optics/photonics by itself does not provide for an easy memory (except for high finesse cavities, or holographic methods), the natural role is for combinatorial logic, storedprogram operation, and analogue or metaphoric information representations. This points to optics being more likely an accelerator or special purpose computer, rather than a multipropose machine.
We are interested finding answer to the question ‘What is the role of optics and photonics’ in computing? In short we believe in bosonification of compute engines. Light are bosons which adhere to the BoseEinstein statistics, which means they are able to occupy the same quantum state. This is unlike Fermions which ‘see’ each other (i.e. Coulomb screening). This natural parallelism has been utilized in fiber optics such as wavelengthdivision multiplexing or more recently optical angular momentum. But are photons superior in terms of energypercompute? The short answer is, when the lightmatterinteraction is enhanced, such as via the Purcell factor (Fp), then electrooptic components are able to be as efficient as modern transistors (~ 10100’s aJ/bit). Our actual goal, however, is to execute a mathematical algorithm by flowing light through a structure. Here the vision is to engineer a photonic structure where data enters and information leaves; i.e. simply flowing light through a system enables computing. Thus, our aims is to synergistically map an algorithm onto photonic hardware. We currently investigate several projects that follow this approach to include a temporal photonic FFT, optical convolution, and NPcomplete oracle, and photonic neuromorphic computer. 
Optical Signal Processing & OFFTThe fast Fourier transform (FFT) is a useful and prevalent algorithm in signal processing. It characterizes the spectral components of a signal, or is used in combination with other operations to perform more complex computations such as filtering, convolution, and correlation. Digital FFTs are limited in speed by the necessity of moving charge within logic gates. An analog temporal FFT in fiber optics has been demonstrated with highest data bandwidth. However, the implementation with discrete fiber optic FFT components is bulky. Here, we present and analyze a design of an optical FFT in Silicon photonics and evaluate its performance with respect to variations in phase and amplitude. We discuss the impact of the deployed devices on the FFT’s transfer function quality as defined by the transmission output power as a function of frequency, detuning phase, optical delay, and loss.
Convolutional neural networks have become an essential element of spatial deep learning systems. In the prevailing architecture, the convolution operation is performed with Fast Fourier Transforms (FFT) electronically in GPUs. The parallelism of GPUs provides an efficiency over CPUs, however both approaches being electronic are bound by the speed and power limits of the interconnect delay inside the circuits. Here we present a silicon photonics based architecture for convolutional neural networks that harnesses the phase property of light to perform FFTs efficiently. Our alloptical FFT is based on nested MachZender Interferometers, directional couplers, and phase shifters, with backend electrooptic modulators for sampling. The FFT delay depends only on the propagation delay of the optical signal through the silicon photonics structures. Designing and analyzing the performance of a convolutional neural network deployed with our onchip optical FFT, we find 100x improvements compared to GPUs when exploring a compounded figureofmerit given by power per convolution over volume. This performance is enabled by mapping the desired mathematical function, an FFT, synergistically onto hardware, in this case optical delay interferometers. 
Optical Neuromorphic ComputingThe longrange interference of optics enables natural convolution and vector matrix multiplication. These can be used in neuromorphic compute systems as shown on the left. A strong experimental thrust to design, build, and demonstrate these proposed systems will serve as a first feasibility proof of using integrated photonics for scalable information processing. Key points of innovation include: 1) arbitrarily scalable Silicon photonic neural network architectures and the testbeds required to demonstrate them, 2) attojoule/MAC electrooptic modulator neurons deploying unitystrong optical index switching materials and plasmonic modes to tailor lightmatter interaction at the nanoscale, and 3) metrics required to simulate and design these systems and benchmarks required to orient them within the broader field of computing applications. The required nonlinearity for the perceptron algorithm is provided via the transfer function of our compact and aJ/bit efficient hybrid integrated photonic/plasmonic modulators. The energy efficiency of this photonic neuromorphic compute engine depends energyperbit function of the modulator neurons. The neural network cascadability from nodetonode is given by the SNRreduction ability of the nonlinear transfer function of the modulator neuron. This novel optical neuromorphic computer could deliver efficiencies of 10^9 GMAC/J, which is beyond the CMOS digital efficiency wall.

Photonic Spiking Neuromorphics for Mirror Symmetry Detection
The ability to rapidly identify symmetry and antisymmetry is an essential attribute of intelligence. Symmetry perception is a central process in human vision and may be key to human 3D visualization. While previous work in understanding neuron symmetry perception has concentrated on the neuron as an integrator, here we show how the coincidence detecting property of the spiking neuron can be used to reveal symmetry density in spatial data. We develop a method for synchronizing symmetryidentifying spiking artificial neural networks to enable layering and feedback in the network. We show a method for building a network capable of identifying symmetry density between sets of data and present a digital logic implementation demonstrating an 8x8 leakyintegrateandfire symmetry detector in a field programmable gate array. Our results show that the efficiencies of spiking neural networks can be harnessed to rapidly identify symmetry in spatial data with applications in image processing, 3D computer vision, and robotics.

Photonic NPComplete Oracle
There are more than 2,000 nondeterministic polynomial time complete problems in our daily lives (e.g. Ising models). We investigate an reconfigurable optical oracle based on cascaded directional couplers in Silicon photonics, for the solution of the Hamiltonian path problem. The optical oracle could solve this NPcomplete problem hundreds of times faster than bruteforce computing. We aim to show that graph theory problems may be easily implemented in integrated photonic networks, down scaling the net size and speeding up execution times. The latter is only determined by the timeofflight of a photon through the network which can lead to 1,000x reduced decision times compared to brutforce electronic computers.

Reconfigurable Optical ComputerThe Reconfigurable Optical Element (ROE) is a novel voltage programmable photonic component utilizing electrooptic effects (i.e. modulation) and optical absorption (i.e. photodetection) to mimic electrical circuit elements to include optical counterparts to R, C, L building blocks. The risktired approach (lowtohigh) is comprised of a) utilizing Silicon photonics foundry processes, b) a hybrid Siliconplasmon design, and c) a metatronicbased design.
By integrating ROEs into an optical mesh network with silicon photonic waveguides, classic problem (e.g. heat transfer problem) can be solved by using the difference in light intensity in the first two approaches and displacement current density in the metatronics approach. Moreover, due to the tunability of the ROE, this reconfigurable optical solver also has the potentials in solving more comprehensive problems such as the Poisson equations, the Diffusion equations and Wave equations. 
Optical Residue ComputingIn the residue number system (RNS) an integer is represented as a set of residues with respect to a set of moduli. Such RNS system can perform arithmetic such as addition, which is executed digitbydigit; a large number is decomposed by factorization using the moduli as basis functions, which allows for parallelism. The builtin parallelism of bosonic photons allow mapping RNSbased arithmetic onto photonic onchip, enabling highly parallel optical compute engines. Here we investigate an photonic RNS adder deploying parallel waveguides, and 2x2 hybrid photonicplasmonic ITO switches. Allowing each switch to be individually reconfigured, via a control signal. This enables a crossbarlike spatial light router, where the residue is represented by spatially shifting the input waveguides relative to the routers outputs. We show that this nanophotonic modulo5 system adder consumes 7.2 fJ/bit, occupying 200 um^2 of footprint, and having a response time of about 5 ps mainly dominated by the drivers setup time. RNS adders could be beneficial for convolution information processing.

Convolution Neural Networks using 4F systemsThe 4F system uses low power laser light to perform convolution functions in parallel at high speed and resolutions, consuming very little power compared to the energy used by current CMOS based technology. It is employed as base unit in CNN, for making intelligent decision, classifications, pattern recognition, in big data challenging problems.
System is based on 2 stages of Fourier Transform (FT). A pattern is formed by shining laser light to an electronically configured Digital Mirror Device DLM 6500, which modulates the laser intensity, coding analog numerical data. At the Fourier plane, the light pattern convolves with a second image and subsequently transformed in the real space. This system can perform Fourier transform calculations 10 times faster than an Nvidia P6000 (100GPixel/s) graphics card (12 TFlops with a Max Power consumption of 250 W), commonly employed for highperformance computation, using just 25 percent of the power. 
Research Overview
Research Tour in OPEN Lab
2D Material Printer
* Video recorded in real time