On-chip bacterial foraging training in silicon photonic circuits for projection-enabled nonlinear classification.

On-chip training remains a challenging issue for photonic devices to implement machine learning algorithms. Most demonstrations only implement inference in photonics for offline-trained neural network models. On the other hand, artificial neural networks are one of the most deployed algorithms, while other machine learning algorithms such as supporting vector machine (SVM) remain unexplored in photonics. Here, inspired by SVM, we propose to implement projection-based classification principle by constructing nonlinear mapping functions in silicon photonic circuits and experimentally demonstrate on-chip bacterial foraging training for this principle to realize single Boolean logics, combinational Boolean logics, and Iris classification with ~96.7 - 98.3 per cent accuracy. This approach can offer comparable performances to artificial neural networks for various benchmarks even with smaller scales and without leveraging traditional activation functions, showing scalability advantage. Natural-intelligence-inspired bacterial foraging offers efficient and robust on-chip training, and this work paves a way for photonic circuits to perform nonlinear classification.

Inter-layer light transition in hybrid III-V/Si waveguides integrated by µ-transfer printing.

We demonstrate low-loss and broadband light transition from III-V functional layers to a Si platform via two-stage adiabatic-crossing coupler waveguides. A 900-µm-long and 2.7-µm-thick III-V film waveguide consisting of a GaInAsP core and InP cladding layers is transferred onto an air-cladding Si photonic chip by the µ-transfer printing (µ-TP) method. An average optical coupling loss per joint of 1.26 dB is obtained in C + L telecommunication bands (1530-1635 nm). The correlation between alignment offset and measured optical coupling loss is discussed with the frequency distribution of µ-TP samples. We also performed a photoluminescence measurement to investigate the material properties in the GaInAsP layer to see if they are distorted by the strong bending stress produced during the pick-up and print steps of the µ-TP process. The peak intensity reduction of 80-90% and a wavelength shift of 0-5 nm (blue shift) were observed after the process. The series of fundamental studies presented here, which combine multiple analyses, contribute to improving our understanding of III-V/Si photonic integration by µ-TP.

Simple and fully CMOS-compatible low-loss fiber coupling structure for a silicon photonics platform.

A simple low-loss fiber coupling structure consisting of a Si inverted-taper waveguide and a 435 nm wide and 290 nm thick SiN waveguide was fabricated with fully complementary metal-oxide semiconductor (CMOS)-compatible processes. The small SiN waveguide can expand to the optical field corresponding to a fiber with a mode-field diameter of 4.1 µm. The fiber-to-chip coupling losses were 0.25 and 0.51 dB/facet for quasi-TE and quasi-TM modes, respectively, at a 1550 nm wavelength. Polarization-dependent losses of the conversion in the Si-to-SiN waveguide transition and the fiber-to-chip coupling were less than 0.3 and 0.5 dB, respectively, in the wavelength range of 1520-1580 nm.

Arbitrary reconfiguration of universal silicon photonic circuits by bacteria foraging algorithm to achieve reconfigurable photonic digital-to-analog conversion.

Reconfigurable/reprogrammable universal silicon photonic circuits represent a paradigm shift in designing photonic devices. However, it is very challenging to perform adaptive arbitrary reconfiguration when the high-dimensional solution of phase distribution cannot be explicitly determined, especially when there are random initial phase errors, which hinder the implementation of novel potential functions in universal circuits. This work presents an arbitrary black-box reconfiguration for universal circuits with random phase errors by a bacteria-foraging algorithm and unlocks a novel function of arbitrary-port-and-arbitrary-bit-resolution reconfigurable 6-bit photonic digital-to-analog conversion. This work offers a general and efficient method to ease multipurpose reconfiguration for universal silicon photonic circuits.

Interferometric autocorrelation of ultrafast optical pulses in silicon sub-micrometer p-i-n waveguides.

We investigated the high-sensitivity interferometric autocorrelation of ultrafast optical pulses utilizing two-photon absorption in sub-micrometer silicon p-i-n waveguides. The autocorrelation sensitivities were evaluated to be about 0.5 and 4.5 × 10-8 W2 for 1- and 0.5-mm devices, respectively. Such sensitivities are about 100 times higher than the traditional two-photon conductivity photodetectors in commercial autocorrelators; thus favor weak pulse characterization. We comprehensively studied the interferometric autocorrelation performances by the experiment and FDTD (finite-difference time-domain) simulation. The pulse energy dependences of measured autocorrelation photocurrents and pulse widths were well explained by the simulation with the free carrier absorption and free carrier plasma effect considered. The autocorrelation error tends to occur if the pulse energy is high enough to cause strong free carrier effects and the threshold pulse energy for error occurrence is increased for shorter devices, but accurate autocorrelation measurement was achieved for sub-Watts pulses at which the influences of free carrier effects on interferometric autocorrelation was negligible. The minimum applicable range of pulse widths was estimated from waveguide dispersion analysis to be ~0.09 and 0.13 ps with a 10% target error for 0.5-mm and 1-mm devices, respectively. The interferometric autocorrelation in sub-micrometer silicon p-i-n waveguides is promising as a monolithic photonic device for on-chip monitor and diagnostics of weak ultrafast pulses.

Silicon traveling-wave Mach-Zehnder modulator under distributed-bias driving.

The silicon traveling-wave (TW) Mach-Zehnder modulator (MZM) is one of the most important devices in silicon photonic transceivers for high-speed optical interconnects. Its phase shifter utilizes carrier depletion of pn diodes for high speed, but suffers low modulation efficiency. Extensive efforts have been made on pre-fabrication optimizations, including waveguides, doping, and electrodes to enhance high-frequency modulation efficiency. Instead, we here propose an adaptive post-fabrication distributed-bias driving method that enables 20%∼30% high-frequency efficiency enhancement at both 10 and 25 Gbps without doing any optimizations for a silicon TW-MZM. This method explores the bias nonlinearity of index modulation which, to the best of our knowledge, is utilized for the first time in driving silicon modulators to improve the efficiency. We demonstrated the viability of this adaptive driving concept to achieve better performance, and this Letter could open new avenues for silicon traveling-wave modulator design and performance trade-off.

High-efficiency strip-loaded waveguide based silicon Mach-Zehnder modulator with vertical p-n junction phase shifter.

We demonstrate a silicon Mach-Zehnder modulator (MZM) based on hydrogenated amorphous silicon (a-Si:H) strip-loaded waveguides on a silicon on insulator (SOI) platform, which can be fabricated by using a complementary metal-oxide semiconductor (CMOS) compatible process without half etching of the SOI layer. Constructing a vertical p-n junction in a flat etchless SOI layer provides superior controllability and uniformity of carrier profiles. Moreover, the waveguide structure based on a thin a-Si:H strip line can be fabricated easily and precisely. Thanks to a large overlap between the depletion region and optical field in the SOI layer with a vertical p-n junction, the MZM provides 0.80- to 1.86-Vcm modulation efficiency and a 12.1- to 16.9-dBV loss-efficiency product, besides guaranteeing a 3-dB bandwidth of about 17 GHz and 28-Gbps high-speed operation. The αVπL is considerably lower than that of conventional high-speed modulators.

Completely CMOS compatible SiN-waveguide-based fiber coupling structure for Si wire waveguides.

For Si wire waveguides, we designed a highly efficient fiber coupling structure consisting of a Si inverted taper waveguide and a CMOS-compatible thin SiN waveguide with an SiO2 spacer inserted between them. By using a small SiN waveguide with a 310 nm-square core, the optical field can be expanded to correspond to a fiber with a 4.0-µm mode field diameter. A coupled waveguide system with the SiN waveguide and Si taper waveguide can provide low-loss and low-polarization-dependent mode conversion. Both losses in fiber-SiN waveguide coupling and SiN-Si waveguide mode conversion are no more than 1 dB in a wide wavelength bandwidth from 1.36 µm to 1.65 µm. Through a detailed analysis of the effective refractive indices in the coupled waveguide system, we can understand mode conversion accurately and also derive guidelines for reducing the polarization dependence and for shortening device length.

Optical autocorrelation performance of silicon wire p-i-n waveguides utilizing the enhanced two-photon absorption.

Optical autocorrelation accuracy was for the first time analyzed for the silicon waveguide based autocorrelators utilizing two-photon absorption (TPA) under various short pulse conditions by numerical simulation. As for autocorrelation operation in the sub-µm silicon p-i-n rib waveguides on the 220 nm SOI (silicon on insulator) wafers, the autocorrelation error of pulse width measurement gradually increases with the increase of the peak power for both Gaussian and hyperbolic secant pulses due to the influence of free-carrier absorption (FCA). For the same pulse type, the relative error is independent of the input pulse width; however different pulse type has different peak power dependency of the accuracy. It was verified that this thin rib waveguide has a TPA responsivity >60 times higher than the thick rib waveguides and the correct pulse width can be measured with a <1% relative error for characterizing ps/sub-ps short pulses of sub-watt peak powers by utilizing the silicon wire p-i-n waveguides as the autocorrelator detector.

Ultra-compact 8 × 8 strictly-non-blocking Si-wire PILOSS switch.

We report on a path-independent insertion-loss (PILOSS) 8 × 8 matrix switch based on Si-wire waveguides, which has a record-small footprint of 3.5 × 2.4 mm2. The PILOSS switch consists of 64 thermooptic Mach-Zehnder (MZ) switches and 49 low-crosstalk intersections. Each of the MZ switches and intersections employs directional couplers, which enable the composition of a low loss PILOSS switch. We demonstrate successful switching of digital-coherent 43-Gbps QPSK signal.