Yiwei Xie, Leimeng Zhuang* and Arthur J. Lowery Picosecond

Nanophotonics 2018; 7(5): 837–852

Research article

Yiwei Xie, Leimeng Zhuang* and Arthur J. Lowery

Picosecond optical pulse processing using a terahertz-bandwidth reconfigurable photonic integrated circuithttps://doi.org/10.1515/nanoph-2017-0113Received November 16, 2017; revised January 23, 2018; accepted February 12, 2018

Abstract: Chip-scale integrated optical signal processors promise to support a multitude of signal processing func-tions with bandwidths beyond the limit of microelectron-ics. Previous research has made great contributions in terms of demonstrating processing functions and device building blocks. Currently, there is a significant interest in providing functional reconfigurability, to match a key advantage of programmable microelectronic processors. To advance this concept, in this work, we experimentally demonstrate a photonic integrated circuit as an optical signal processor with an unprecedented combination of two key features: reconfigurability and terahertz band-width. These features enable a variety of processing func-tions on picosecond optical pulses using a single device. In the experiment, we successfully verified clock rate mul-tiplication, arbitrary waveform generation, discretely and continuously tunable delays, multi-path combining and bit-pattern recognition for 1.2-ps-duration optical pulses at 1550 nm. These results and selected head-to-head com-parisons with commercially available devices show our device to be a flexible integrated platform for ultrahigh-bandwidth optical signal processing and point toward a wide range of applications for telecommunications and beyond.

Keywords: integrated optics; photonic integrated circuit; picosecond pulse processing; waveguide.

1 IntroductionOptical pulses are essential in many optical systems and have applications across different fields such as telecom-munications, sensing, imaging and metrology [1–3]. The processing of optical pulses shares some commonalities with digital signal processing of electrical samples. A variety of optical pulse processing functions can therefore be synthesized by different combinations of optical com-ponents such as splitters, combiners and delay lines [4, 5].

The advance in photonic integrated circuit (PIC) tech-nologies has enabled the implementation miniaturization of optical components [6–16]. PICs can be controlled accu-rately and are stable. Regarding integrated optical signal processors using PICs, previous research has made great contributions in terms of circuit design and implementing various processing functions, with a wide range of appli-cations having been demonstrated [17–24]. The constant development of essential components and building blocks such as splitters(combiners) [25, 26], (de)multiplexers [23, 27], couplers [4, 7], delay lines [8, 28] and phase shifters [16, 22] have shown promising results of performance in terms of size, loss, bandwidth, coupling efficiency, control power consumption and fabrication accuracy. On the circuit level, a well-studied circuit topology is the arrayed waveguide grating router (AWGR) [29], which joins two slab couplers with an array of linearly incremental wave-guide delays. AWGRs are able to perform M × N Fourier transforms, and therefore, their use has been investigated in optical communication and spectroscopic systems [30, 31]. While AWGRs can be used as building blocks in more complex systems [32–35], their rigid topology limits their functional flexibility. In contrast, topologies compris-ing coupler arrays and delay lines offer design flexibil-ity and easy incorporation of tuning elements [36–39]. A key advantage for signal processing is that these features support custom-synthesis of processing responses in both time and frequency domains, i.e. impulse responses and spectral responses. To date, a number of important func-tions have been demonstrated using such topologies,

*Corresponding author: Leimeng Zhuang, Electro-Photonics Laboratory, Department of Electrical and Computer Systems Engineering, Monash University, Clayton, VIC3800, Australia, e-mail: [email protected]. http://orcid.org/0000-0002-5054-4917Yiwei Xie and Arthur J. Lowery: Electro-Photonics Laboratory, Department of Electrical and Computer Systems Engineering, Monash University, Clayton, VIC3800, Australia

Open Access. © 2018 Leimeng Zhuang et al., published by De Gruyter. This work is licensed under the Creative Commons Attribution- NonCommercial-NoDerivatives 4.0 License. Brought to you by | Monash University - Caul

AuthenticatedDownload Date | 6/4/18 7:14 AM

https://doi.org/10.1515/nanoph-2017-0113

mailto:[email protected]

http://orcid.org/0000-0002-5054-4917

http://orcid.org/0000-0002-5054-4917

838 Y. Xie et al.: Picosecond optical pulse processing using a terahertz-bandwidth reconfigurable PIC

including delay lines [40–42], temporal waveform shapers [24, 43], spectral filters [44–51], multi-port switches [52–56], Fourier transforms [23, 45], correlators [57], beam-formers [58–61], quantum processors [62, 63] and mesh networks [64–66].

Although it has explored a wide diversity of func-tions, most of the previous research was directed toward engineering such devices for one particular function. This paradigm is inflexible, costly and risky, which slows their commercial adoption. It is therefore highly desirable for a photonic chip to be able to switch between different func-tions by means of programming and also can be tuned and optimized for post-production. This not only provides a path to significantly reducing the cost per function, but more importantly, one common verified design can be used for many applications [65, 67–69]. This concept has been successfully implemented in microelectron-ics, such as FPGAs and programmable digital processors [70]. The future adoption of PICs obviously also depends on whether they can provide adequate performance. In particular, the processing of picosecond optical pulses requires terahertz-processing bandwidths to guarantee the quality of the output signals. However, it is challeng-ing to provide all these device features on a small chip area. High-index-contrast waveguides such as silicon on insulator and III-V semiconductors support very small bend radii for compact designs [7, 8]. However, currently, they still have significant propagation losses in the order of a few decibels per centimeter and a variation of propa-gation constant with wavelength. In addition, these wave-guides also have strong nonlinear effects at high optical powers (e.g. >100 mW) [71, 72], which are detrimental to pulse processing. On the other hand, low-index-contrast

waveguides such as silicon dioxide and silicon oxynitride provide good performance, but typically associate with larger device sizes [9, 12].

Here, we present a first demonstration of implement-ing a diversity of terahertz-bandwidth processing opera-tions on picosecond optical pulses using an optical signal processor chip. This includes two binary-tree coupler arrays and an array of 16 delay lines with 6.25-ps differen-tial delay. It is fabricated in stoichiometric silicon nitride Si3N4/SiO2 waveguides on an area of 0.9 cm2. The fiber-to-fiber insertion loss of the packaged chip is 5 dB at optimal coupling. In the experiments, we have verified that our processor chip is able to implement multiple and distinc-tively different processing functions, including terahertz-bandwidth clock rate multiplication, arbitrary waveform generation, discretely and continuously tunable delays, multi-path combining and bit-pattern recognition, for 1.2-ps-duration optical pulses at 1550 nm.

2 Device descriptionFigure 1 shows the optical signal processor chip fabricated using a Si3N4/SiO2 waveguides [10, 14] (TriPleX, a propri-etary waveguide technology of LioniX International). The design comprises three serial concatenated sections as depicted in Figure 1A: a binary-tree-structured 8 × 16 coupler array based on 2 × 2 tunable Mach-Zehnder inter-ferometers (MZIs), a 16-arm array of linearly incremen-tal delay lines with each arm equipped with a tunable phase shifter and an inter-arm delay interval of 6.25 ps and a binary-tree 16 × 8 coupler based on 3-dB directional

Figure 1: (A) Schematic of the 16-arm tapped delay line filter; (B) fabricated device (inset: cross section of the waveguide structure); (C) packaged chip.

Brought to you by | Monash University - CaulAuthenticated

Download Date | 6/4/18 7:14 AM

Y. Xie et al.: Picosecond optical pulse processing using a terahertz-bandwidth reconfigurable PIC 839

couplers. The tuning elements are implemented thermo-optically using chromium heaters. This architecture minimizes the circuit complexity and heat dissipation compared with designs using a parallel array of couplers (which require more building blocks and tuning ele-ments), while guaranteeing full control of both amplitude and phase of each optical path (i.e. full control of complex coefficients, as in a spatial light modulator [73]). The fabri-cated chip is shown in Figure 1B, which has a chip area of 0.9 cm2. As shown in the inset of Figure 1B, the waveguide uses a “double-stripe” geometry, i.e. two stripes of Si3N4 with a width of 1.2 μm, a thickness of 170 nm and a vertical spacing of 500 nm filled with SiO2. This waveguide has a reported propagation loss of 0.15 dB/cm, group index of 1.71 and wavelength dependency of 2 × 10−5/nm [10, 14]. It supports a single mode at 1550 nm and is optimized for light coupling with TE polarization. The fully packaged chip is shown in Figure 1C, where all the heaters are wire-bonded to a carrier printed circuit board and optical I/Os are pigtailed with polarization-maintaining fibers.

More specifically, Figure 2 shows the chip layout. We used OptoDesigner (PhoeniX Software) to layout the chip. The design is a compromise between chip area, possible inter-component cross-talk and efficient routing of both optical and electrical paths. An S-shaped spiral with a bend radius of 150 μm is used for the delay lines to mini-mize the area. The 4-level binary-tree-structured coupler arrays are folded into a two-column layout. This layout guarantees two features for proper operation of the pro-cessor chip: (1) that all the couplers from the same stage of the binary tree are connected with equal lengths of wave-guide to the other stages, so that the delay differences between optical paths in the circuit are only determined

by the designated delay lines, and (2) that the tunable cou-plers are equally spaced in each column with a minimum spacing of 250 μm between a heater and a neighboring waveguide to minimize the thermal cross-talk between them. The heaters in the phase shifter section are equally apart with a spacing of 350 μm and are placed outside the delay spirals to minimize their impact on the areas dense with waveguides. Regarding the electrical paths, several considerations are taken into account. Two inter-connected ground tracks with widths of 300 μm are used such that all heaters are able to connect to a ground track with a short lead so as to minimize electrical cross-talk as well as the current per ground track. The track width is 100 μm, to accommodate the current required for the full tuning range of each heater (each heater has 600 Ω resistance). Moreover, the circuit topology has test inputs and outputs, allowing access to different circuit sections independently, which simplifies device characterization and increases the operational flexibility.

3 Experimental results

3.1 Basic characterization

For the characterization of the circuit components, we chose the section between input 8 and output 8, as shown in Figure 1A, which comprises a simple bar-port delay interferometer with two arms (arms 1 and 2). The measured optical power responses at different heater biases of arm 2 are shown in Figure 3A. A bias power of 0.46 W is needed to generate a phase change of 2π in arm

Figure 2: Chip layout.




2, resulting in a frequency shift of the optical response by an entire frequency spectral range (FSR). Figure 3B shows the extinction ratio (ER) as a function of the bias of the tunable coupler (MZI 15). The resulting ER is about 23 dB. The effects of thermal cross-talk are shown in Figure 3C and D. It is evident in Figure 3C that the null depth is insensitive to the cross-talk from the neighbor-ing heaters (MZI 14 and MZI 7). Figure 3D shows the effect of cross-talk between heaters in arms 2 and 3 on arm 1, which results in an additional phase change and mani-fests as a frequency shift of the optical response. Owing to a sufficient inter-arm spacing in our design, the cross-talk effect is only about 3% between two neighboring arms and almost negligible for larger spacing. In prac-tice, this effect can be compensated using the heater on the target arm.

For the control of the tuning elements, a custom 12-bit multi-channel voltage controller was used (LioniX Inter-national). In principle, automatic-control algorithms can be developed to drive the voltage controller for realizing automatic control, e.g. automatic translation from target functions into tuning element parameter settings [74]. However, in this work, the control for chip characteriza-tion and reconfiguration were performed manually with the parameter settings recorded in a look-up table. The difficulty of generating an auto-control algorithm lies

mainly in the characterization of the initial offsets, tuning function to input and cross-talk between the tuning ele-ments, which constitute a transfer matrix with the dimen-sion of N × N, with N being the number of tuning elements on chip.

3.2 Terahertz-bandwidth optical clock rate multiplication

To date, integrated pulse train sources have been widely investigated in both academia and industry. For practical applications, switchable repetition rates are a desirable feature. However, this is inherently difficult to implement in many pulse sources due to a fixed cavity length. An effective way to overcome this drawback is to multiply the pulse repetition rate externally by means of modifying the optical spectrum of the pulse train signal using an optical filter. It is also desirable to implement such filters on chip with the lasing itself.

In this context, our processor chip can be configured to be a tapped delay line filter with a maximum of 16 taps, where middle input 4 and output 4 are used as optical I/Os. Mathematically, such a filter can be modeled as a digital finite impulse response filter [4], the transfer func-tion of which is given by

Figure 3: Characterization results of the sections between input 8 and output 8 (MZI 15 connected with arms 1 and 2): (A) spectral shift when arm 2 has DC power; (B) ER and phase change of MZI 15; (C) thermal cross-talk-induced spectra null depth variation (cross-talk from MZI 14 and MZI 7); (D) thermal cross-talk-induced phase change (cross-talk from arms 3 and 4) with respect to the applied DC power.




1 j j20

( ) nN fn

nnH f a e e

Nϕ π τΓ − − −

= = ∑

(1)

where f is the optical frequency, Γ is an amplitude nor-malization constant, N is the number of taps (arms) and nτ is the delay of tap n. For the nth tap, the amplitude and phase coefficients are an and ϕn, respectively. These two coefficients can be varied by means of the tunable MZIs and phase shifters. This also provides on-off switching of each filter arm and enables filter passband shaping as well as frequency tuning.

Figure 4 shows a number of measured filter responses with respect to different coefficients. We used a Luna OVA5000, and a stopband null suppression of about 30 dB was achieved across the entire bandwidth of interest. The maximum FSR of the filter is determined by the inter-arm delay interval τ = 6.25 ps by the relation fFSR = 1/τ = 160 GHz. Figure 4A shows the filter response with all arms switched fully on (an = 1) and phase-aligned (ϕj = ϕk). A sinc response is expected. The relative error between the experimental and the theoretical response using Eq. (1) is around 5% due to the limited ER of the

MZI couplers on the chip. Figure 4B demonstrates the frequency shifting of the filter passband without chang-ing the passband shape. This was performed across a full spectral range, by adding a linearly incremental phase shift across all arms. Figure 4C and D demonstrates the tuning of the passband bandwidth and FSR. This was performed by tuning the amplitude coefficients of the filter, implemented by means of the tunable MZIs in the 1 × 16 splitting circuit. The corresponding filter coeffi-cient configurations are specified alongside for clarity. The capability of defining both the passband repeti-tion bandwidth and the frequency spacing between the stopband nulls is the key for implementing optical clock rate multiplication, where the filter passbands select the desired frequency components of the clock signal and the stopband nulls provide the maximum suppression to the undesired ones.

To verify the optical clock rate multiplication func-tion [75], we configured our processor chip into two dif-ferent tapped delay line filter topologies: (A, B) a 2-tap filter with a FSR of 20 GHz and (C, D) a 4-tap filter with a FSR of 40 GHz, by selectively switching on a number of

Figure 4: Filter frequency response measurements: (A) comparison between measurement and calculation; (B‒D) demonstrations of pass-band frequency shifting, passband shaping and variation of filter’s free spectral range (inset table: filter configurations).




arms. The measured filter responses optimized for equal-suppression stopband nulls are shown in Figure 5B and D, respectively. Figure 5A and C shows the spacing between

the passband peaks. The measurements were performed with a bandwidth about 2.5 THz. A FSR variation smaller than 3% and stopband-null suppression larger than 30 dB

Figure 5: Measured filter responses and FSR uniformity across a bandwidth of 2.5 THz: (A, B) for a 2-tap 20-GHz-FSR filter topology; (C, D) for a 4-tap 40-GHz-FSR filter topology. Measured signal spectra and waveform obtained from a PD: (E, F) input clock signal at 10 Gpulses/s; (G, H) output clock signal at 20 Gpulses/s; (I, J) output clock signal at 40 Gpulses/s.




were achieved across this bandwidth. Then, we applied an optical clock/periodic pulse train signal with a repeti-tion rate of 10 Gpulses/s as the input. The corresponding optical spectrum and waveform obtained by a 70-GHz bandwidth photodetector (PD) and 62-GHz bandwidth real-time scope are shown in Figure 5E and F, respectively. The observed pulse width is about 20 ps. The filter output using the 2-tap 20-GHz-FSR filter topology is shown in Figure 5G and H, showing the spectrum and waveform of the output clock/pulse train signal, respectively. Evi-dently, the filter suppresses every other comb line in the spectrum, with a suppression ratio of 29 dB, and thereby changes the comb line spacing from 10 to 20 GHz. Con-currently, the pulse interval in the time domain reduces from 100 to 50 ps. This result shows an implementation of a doubling of the pulse repetition rate. The filter output using the 4-tap 40-GHz-FSR filter topology is shown in Figure 5I and J. Likewise, this result shows a quadrupling of the pulse repetition rate, where the filter suppresses three in every four spectral comb lines, resulting in a comb line spacing of 40 GHz and correspondingly a pulse inter-val of 25 ps. These demonstrations provide verifications of the function of a reconfigurable optical clock multiplier.

To further demonstrate the qualities of our output clock signal, we applied it in a PRBS-signal transmis-sion experiment and made a head-to-head comparison with a reference implementation using a commercial optical clock multiplier based on free-space optics (PriTel optical clock multiplier). Using the chip and the commer-cial multiplier, respectively, we first generated 20- and 40-Gpulses/s clock signals from a seed clock signal at 10 Gpulses/s with a pulse duration of 1.2 ps, where both devices were optimized to provide maximum suppression to the undesired frequency components in the outputs. The outputs of the clock multipliers are fed to a 35-Gbits/s IQ modulator (TeraXion IQM, only one modulator is used), which are driven by a pulse pattern generator (SHF BPG

44E). For a clear comparison, Figure 6 shows the electri-cal spectra corresponding to their waveforms after detec-tion. For both the 20- and 40-Gpulses/s cases, the filter chip demonstrates similar performance to the commercial multiplier, showing similar sidelobe suppression ratios, i.e. 35 dB for the 20-Gpulses/s case and 30 dB for the 40-Gpulses/s case. Then, we used these generated clock signals as the light source in a back-to-back transmission system. We externally modulated them with on-off keying with the symbol rates matched with the pulse repetition rates. As a performance indicator, the bit error rates (BER) versus optical signal-to-noise ratio (OSNR) of the received on-off keying signals were measured and compared, as shown in Figure 7A. The filter chip has a nearly identical BER performance to the commercial multiplier for both 20 and 40 Gpulses/s. To show the utility of these signals, Figure 7B also includes an optimal transmission case where the seed clock signal was generated using a com-mercial mode-locked laser (MLL) (Lumentum ERGO) at 10 Gpulses/s as a pulse source into the modulator. As shown in Figure 7B, the filter chip shows nearly no degradation to the signal.

3.3 Arbitrary waveform generation

Defining signal waveforms at high speeds is essential for modern telecommunications and radar systems, where large signal bandwidths are used. Conventional implementations of high-speed arbitrary waveform gen-erators (AWG) require fast and power-hungry digital-to-analog convertors. As the demand for signal bandwidth increases, it is challenging to provide all-electronic solu-tions. A state-of-the-art commercial AWG with a sampling rate about 100 GSa/s and an electrical bandwidth about 40 GHz is still too expensive to be adopted for applications out of the laboratory. However, photonic approaches open

Figure 6: Spectral characteristics of the output clock signals generated using the filter chip and commercial multiplier: (A) for the output at 20 Gpulses/s; (B) for the output at 40 Gpulses/s. Inset: OOK eye diagrams comparison between the outputs of the filter chip and the com-mercial reference when the OSNR = 15 dB.




an alternative path for implementing high-speed AWGs with potential for low cost and power consumption.

To date, a number of photonic chip implementations of AWG have been reported with promising results [18, 32, 42]. One straight-forward way to synthesize a particular electrical waveform is to use the time-domain impulse response of a tapped delay line filter. In this approach, the impulse response of the filter converts a short input pulse into a serial sequence of multiple pulses with their intensity envelop defined by the amplitude coefficient of each tap. Then, an according electrical waveform can be obtained by detecting this optical signal. As explained earlier, our processor chip can be configured to be a 16-tap delay line filter with a time-domain impulse response res-olution of 6.25 ps, equivalent to an AWG sample rate of 160 GSa/s.

Figure 8 shows the synthesis of a number of repre-sentative waveforms from an input pulse with a duration of 2 ps, verifying the capability of arbitrary waveform gen-eration. The corresponding filter coefficient configura-tions are specified for clarity. Figure 8A and B shows the waveforms of a falling ramp and a rising ramp, respec-tively. Figure 8C shows a triangular waveform. Figure 8D is a waveform showing height control of ramps and steps. Figure 8E shows a waveform having both climbing and falling steps. Figure 8F is a waveform approximating a sinc shape. Figure 8G is an approximation of a square waveform. Figure 8H shows a waveform of a random binary bit pattern. The filter configurations are shown in the inset table. These results provide a demonstration of the AWG with the waveforms directly synthesized in the time domain. In these results, the relative error between the simulation and theoretical results is less than 8%, the deviations from the ideal waveforms are mainly caused by six limiting factors: limited ER of the MZI couplers on the chip, limited bandwidths of PD and oscilloscope, limited noise floor of the optical amplifier, limited ER and finite

time duration of the input optical pulses. In practice, these limitations should be taken into account for system design in accordance to the performance requirement. Some pos-sible measures for improvement include, for example: using multi-stage MZI architectures with higher coupler ERs [56], reducing chip insertion loss [12], designing proper sample rate based on the electrical bandwidth, and using narrower and cleaner optical pulses as input [76].

3.4 Discretely and continuously tunable delays

Tunable delays are an essential building block for optical signal processors that perform signal timing control [42, 77]. Our processor chip is able to implement two types of tunable delay lines. Type 1 is with digitally switched tun-ability and a discrete delay steps. The other type, i.e. type 2, is with continuous variable delay.

For the implementation of type 1 delay line, the switch-ing matrices on the chip perform as an end-to-end optical path selector routing the signal from the chip input (input 4 or 5) to output (output 4 or 5) with one of the 16 delays in its path. As such, the resulting delay line enjoys the terahertz-level bandwidth of the waveguide and features a delay tuning range from 0 to 93.75 ps, with a tuning step of 6.25 ps. Figure 9 shows the measured delay responses in the frequency domain, group delay (GD) and an experi-mental verification of its applicability to a narrow optical pulse with a duration of 1.2 ps, i.e. a bandwidth of 0.4 THz.

For the implementation of the type 2 delay line, the chip is configured to be an asymmetric MZI (A-MZI) com-prising two adjacent delay spirals (e.g. arms 1 and 2), equivalent to a 2-tap delay line filter as explained earlier. In this case, the filter passband provides a delay value between those of the two spirals. This delay value can be continuously tuned by means of controlling the coupler

Figure 7: BER measurements: (A) received on-off keying signals for the commercial optical clock multiplier and chip; (B) optimal transmis-sion case for MLL and chip with different pulse repetition.




in the A-MZI. However, as a trade-off with the continu-ous tunability, such a delay implementation has a band-width limitation as shown in Figure 10A, i.e. half of the filter FSR, which in our case is 80 GHz. Figure 10A and

B illustrates the spectra and GD of different delay cases. Figure 10C shows the measured delay responses as a type 2 delay line and its corresponding timing control on a square pulse with a duration of 100 ps, equivalent to a pulse bandwidth about 10 GHz, which verifies its function. In a further experiment, the eye diagrams of a 10-Gbaud on-off-keying signal passing through the delay line was measured and shown in Figure 10D. It can be seen that the signal has clear eye opening over the entire delay tuning range of 93.75 ps, verifying the device performance uni-formity and its applicability for optical communications. The delay value has little effect on the time jitter, showing an insignificant increase of 0.5% of the signal baud rate between the maximum minimum delays. This increase is mainly attributed to a higher ASE noise incurred during optical amplification.

3.5 Multi-path combining (optical time division multiplexing (OTDM))

Combining multiple inputs enables generation of a signal streams with high symbol rates by combining a number of signal tributaries with lower symbol rates. This technique benefits the transmission spectral efficiency and relaxes the electrical bandwidth requirement on modulators [78].

To demonstrate the multi-path combining function using our processor chip, we configured the input coupler matrix to be an array of four parallel equal-length routing paths, which then connect, respectively to four delay spirals with delay increments of 25 ps as shown in Figure 11A. After the delay section, the coupler matrix performs the combin-ing and leads the combined signal to the middle output. This chip configuration is able to multiplex four pulse trains of 10 Gpulse/s into a 40-Gpulse/s stream. In the experiment, we split a 10-Gpulse/s, 1.2-ps-duration pulse train into four equal portions and time-aligned them at the chip inputs. The chip output was detected, and the resulting electrical wave-form is shown in Figure 11B. This result shows a pulse train with a pulse interval of 25 ps, i.e. a pulse rate of 40 Gpulse/s, which verifies the combining function as designed.

3.6 Applications for optical communications

3.6.1 Nyquist-OTDM

“Orthogonal” time division multiplexing (OrthTDM) is based on interleaving M modulated sinc pulses per TDM symbol. A practical application is to create a high-rate optical channel using lower-rate optical modulators [78].

Figure 8: Measured waveforms (blue solid line, the waveforms are normalized to the peak intensity) from the filter: (A) falling ramp; (B) rising ramp; (C) triangular waveform; (D) waveform having ramps and steps; (E) waveform having both climbing and falling steps; (F) sinc-function waveform; (G) square waveform and (H) binary bit pattern waveform. The black dashed lines outline the envelopes of the ideal waveforms. Inset table: filter configurations.




In [79], we proposed a direct implementation of a near-Nyquist WDM filter, using an FIR with amplitude weight-ing of each of its waveguides to give a truncated-sinc (TS) impulse response. The filter is fed with OTDM pulses, and the output is Nyquist-OTDM signal. In this case, the recon-figurable filter is working as a FIR filter, considering the FSR of the filter and receiver speed limitation, we tuned the filter to have a near rectangular frequency response, with a nominal bandwidth, B, of 40 GHz and its impulse response is a truncated sinc with pulse width 25 ps, which can be expressed as [79]

MLL

MLL MLL/ 2( ) sinc rect ,

2T

t T Th t t

T∆

−= ⋅ −

(2)

where the rectangular function used here is

MLL

MLL1 [0 ]rect ( ) ,

0 otherwiseT

t Tt

≤ ≤ =

(3)

TMLL is the repetition rate of the MLL and ΔT is the pulse width of the truncated sinc. The reconfigurable filters have 16 arms with differential delays of Δt = 6.25 ps, giving a FSR of 160 GHz. The power in each arm of the filter can be calculated as

21 sinc ( 1) / 1 16.

2MLL

n

Ta n t T n

A∆ ∆

= − − ≤ ≤

(4)

The FIR filter’s frequency response and waveform of the FIR filter with an input pulse with a duration of 1.2 ps are shown in Figure 12A and B. This filter is fed with OTDM pulses created by splitting, modulating, delaying then com-bining pulses from a MLL. Here, the MLL produces a train of short-pulses at 100-ps intervals and modulated with OOK at 10 Gbaud by an intensity modulator. The spectra before and after the chip are shown in Figure 12C; the modulated signals are shaped into near rectangular shapes by the FIR filter.

Figure 9: The measured delay responses in the frequency domain, group delay (GD) and an experimental verification of its applicability to a narrow optical pulse with a duration of 1.2 ps, i.e. a bandwidth of 0.4 THz.(A) Frequency responses and (B) group delays of the programmable filter when tuning the GD ranges from 0 to 93.75 ps with a tuning step of 6.25 ps. (C) Impulse responses of the filter for different delay cases when the input is 1.2-ps pulses for 16 GD cases.




At the receiver, the Nyquist-OTDM signal is ampli-fied and filtered by an 80-GHz Gaussian 1st-order BPF to remove energy from the higher-order passbands of the FIR filter. Figure 12D shows the signal quality for Nyquist-OTDM system when changing the OSNR and an open eye diagram at OSNR of 15 dB.

3.6.2 Optical bit pattern recognition

We further validate the operation of the correlator filter [57] by demonstrating all-optical recognition of 4-bit binary

phase-shift-keyed (BPSK) patterns. This could be used to detect training identities, for example. Figure 13A shows the optical intensity waveform of a BPSK pattern input to the correlator. The input consists of four consecutive pulses at 40 Gpulse/s, with equivalently a delay interval of 25 ps between the pulses. The pulses are generated from a CW external cavity lasers (ECLs) and BPSK encoding is achieved by a modulator (TeraXion IQM, only one port is used). The phases of the input signal are encoded as [0, π, 0, π]. Our filter is configured such that only four arms are in use with their delay interval, amplitude and phase coefficients matching the input BPSK data amplitude.

Figure 10: (A) Frequency responses and (B) group delays of the programmable filter when tuning the coupling ratio, ҡ of the A-MZI. (C) Impulse responses of the filter when the input is 100-ps pulse for different coupling ratio cases and their corresponding OOK eye diagram of a 10-Gbps PRBS signal. Note: Case 1 (blue): ҡ = 0.4, case 2 (yellow): ҡ = 0.5, case 3 (orange): ҡ = 0.6, case 4 (purple): only arm 16 has power.

Figure 11: (A) Principle of multi-path combiner (optical time division multiplexing) realization. (B) OTDM signal realized by inputting a 1.2-ps pulse trains into four ports of the chip.




Figure 13B–F are the auto-correlation signal of the BPSK signal with the correlator tuned to have phases of [0, π, 0, π], [0, 0, 0, 0], [0, π, π, π], [π, π, 0, π] and [0, 0, π, π]. Here, the heights of the peaks, normalized with respect to the maximum auto-correlation peak in Figure 13B. We observe that the filter is capable of distinguishing the different pat-terns. The auto-correlation signals show the same features as the theoretically expected values (red dashed lines), with a relative error of around 7% because of the limited ER of the MZI couplers on the chip and limited bandwidths of PD and oscilloscope; further validating the correlator function. It is worth mentioning that in principle our filter design can recognize many other patterns, e.g. ASK and QPSK, on a maximum of 16 pulses spaced by 6.25 ps because of the full control of filter coefficients (i.e. amplitude and phase) in all 16 arms. However, the demonstration of this is beyond the speed limits of our measurement equipment.

4 Discussions and conclusionsWe have shown a pathway toward the exciting prospect of reconfigurable optical signal processor chips. The PIC

implementation also facilitates such devices to be inte-grated by means of micro-assembly with active circuit components such as lasers, modulators and photodiodes in one unit [80], opening a possibility for creating fully integrated, multi-purpose optical processing “engine” as a stand-alone device. Such processor chips enable variations of processing functions by means of electrical control, which features high accuracy as well as robustness, and implies potential for a wide range of applications.

The terahertz-bandwidth and low-loss properties of the silicon nitride waveguide allow processing of pico-second optical pulses as good as commercial counter-parts based on free-space optics. Considering a future direction of development, ultrahigh-bandwidth process-ing on femtosecond pulses would open a new category of applications. In principle, the silicon nitride waveguide used in our work supports a transparent window from UV to near IR and has lower wavelength-dependency in terms of loss than silicon waveguide [14]. However, as an important performance aspect, signal processors based on interferometric circuits have their spectral repeat-ability (FSR uniformity) depending on the waveguide index variation and circuit parameter accuracy. The

Figure 12: (A) Frequency response and (B) impulse response of the FIR filter. (C) Spectra of the modulated OTDM signal before and after the FIR filter. (D) BER measurements of the received on-off keying signals. Light-colored solid lines in Figure 12A and B are ideal spectra and impulse response, respectively.




direction of effort for improvement lies mainly in the physical layer design of the waveguide and fabrication accuracy as well as uniformity, which requires in itself long-term development. On the circuit level, a chal-lenge for the topology design is the delay accuracy of the optical paths in the processor. This would be a diffi-cult task for ultrahigh-bandwidth applications, because different types of building blocks such as straights, bends, couplers and tuning elements may be included

in different paths and each may introduce a waveguide index variation. Three practical methods may be consid-ered to minimize the effect. One way is to use uniform layout of delay lines and introduce required delay dif-ferences only in the straights. The second is to using binary-tree coupler topology to implement equal-length splitters (combiners). The last is to try to route all delay lines across as much as possible the same regions on the wafer instead of distributed in different regions.

Figure 13: (A) Input signal amplitude with bit pattern [0, π, 0, π]. (B–F) Correlator outputs with correlator bit patterns (B) [0, π, 0, π]; (C) [0, 0, 0, 0]; (D) [0, π, π, π]; (E) [π, π, 0, π] and (F) [0, 0, π, π].




This work demonstrates reconfigurability using thermo-optical tuning mechanism. Further device develop-ment could consider electro-optical [11, 13], graphene-based [6] and stress-optical tuning mechanism [81] to lower power consumption and increasing tuning speed. Moreover, our circuit topology is also implementable in other waveguide materials. This means that further development may con-sider increasing device compactness as well as power efficiency and even the possibility of hybrid electronics-photonics integration using silicon technology platforms. Alternatively, monolithic integration of both passive and active circuit components using III-V semiconductor techno-logy platforms would also bring desirable features and new capabilities to such devices. The demonstrated functions and comparisons with the commercially available devices underline not only versatile and high-performance manip-ulation of optical short pulses on the chip scale, but more significantly, a flexible integrated device platform for ultra-high-bandwidth optical signal processing.

Acknowledgments: This work is funded by the Austral-ian Research Council under grants FL130100041 and CE110001018. We thank LioniX International, The Neth-erlands, for fabricating and packaging the test chip. We thank Ms Caterina Taddei with the Laser Physics and Non-linear Optics (LPNO) group at the University of Twente, The Netherlands, for the chip quality check. We thank Binhuang Song from Monash University for the assistance in the experiments.

References[1] Soto MA, Alem M, Shoaie MA, et al. Optical sinc-shaped Nyquist

pulses of exceptional quality. Nat Commun 2013;4:2898.[2] Udem T, Holzwarth R, Hänsch TW. Optical frequency metrology.

Nature 2002;416:233–7.[3] Liu J, Dai J, Chin SL, Zhang XC. Broadband terahertz wave remote

sensing using coherent manipulation of fluorescence from asymmetrically ionized gases. Nat Photon 2010;4:627–31.

[4] Madsen CK, Zhao JH. Optical filter design and analysis: a signal processing approach, 1st ed. New York, USA, Wiley, 1999.

[5] Lowery AJ, Zhuang L, Corcoran B, Zhu C, Xie Y. Photonic circuit topologies for optical OFDM and Nyquist WDM. J Lightwave Technol 2017;35:781–91.

[6] Liu M, Yin X, Ulin-Avila E, et al. A graphene-based broadband optical modulator. Nature 2011;474:64–7.

[7] Smit M, van der Tol J, Hill M. Moore’s law in photonics. Laser Photon Rev 2012;6:1–13.

[8] Bogaerts W, De Heyn P, Van Vaerenbergh T, et al. Silicon micror-ing resonators. Laser Photon Rev 2012;6:47–73.

[9] Moss DJ, Morandotti R, Gaeta AL, Lipson M. New CMOS-compat-ible platforms based on silicon nitride and Hydex for nonlinear optics. Nat Photon 2013;7:597–607.

[10] Roeloffzen CGH, Zhuang L, Taddei C, et al. Silicon nitride microwave photonic circuits. Opt Express 2013;21:22937–61.

[11] Leuthold J, Koos C, Freude W, et al. Silicon-organic hybrid electro-optical devices. J Sel Topic Quantum Electron 2013;19:114–26.

[12] Heck MJR, Bauters JF, Davenport ML, Spencer DT, Bowers JE. Ultra-low loss waveguide platform and its integration with silicon photonics. Laser Photon Rev 2014;8:667–86.

[13] Melikyan A, Alloatti L, Muslija A, et al. High-speed plasmonic phase modulators. Nat Photon 2014;8:229–33.

[14] Wörhoff K, Heideman RG, Leinse A, Hoekman M. Triplex: a versatile dielectric photonic platform. Adv Opt Technol 2015;4:189–207.

[15] Morrison B, Casas-Bedoya A, Ren G, et al. Compact brillouin devices through hybrid integration on silicon. Optica 2017;4:847–54.

[16] Xie Y, Geng Z, Zhuang L, et al. Programmable optical processor chips: towards photonic RF filters with DSP-level flexibility and MHz-band selectivity. Nanophotonics 2017;7:421–54.

[17] Levy JS, Gondarenko A, Foster MA, Turner-Foster AC, Gaeta AL, Lipson M. CMOS-compatible multiple-wavelength oscillator for on-chip optical interconnects. Nat Photon 2010;4:37–40.

[18] Khan MH, Shen H, Xuan Y, et al. Ultrabroad-bandwidth arbitrary radiofrequency waveform generation with a silicon photonic chip-based spectral shaper. Nat Photon 2010;4:117–22.

[19] Norberg EJ, Guzzon RS, Parker JS, Johansson LA, Coldren LA. Programmable photonic microwave filters monolithically integrated in InP-InGaAsP. J Lightwave Technol 2011;29:1611–9.

[20] Kippenberg TJ, Holzwarth R, Diddams SA. Microresonator-based optical frequency combs. Science 2011;332:555–9.

[21] Morichetti F, Ferrari C, Canciamilla A, Melloni A. The first decade of coupled resonator optical waveguides: bringing slow light to applications. Laser Photon Rev 2012;6:74–96.

[22] Marpaung D, Roeloffzen CGH, Heideman R, Leinse A, Sales S, Capmany J. Integrated microwave photonics. Laser Photon Rev 2013;7:506–38.

[23] Goh T, Itoh M, Yamazaki H, Saida T, Hashimoto T. Optical Nyquist-filtering multi/demultiplexer with PLC for 1-Tb/s class super-channel transceiver. In: Optical Fiber Communication Conference (OFC), Los Angeles, USA, 2015:Tu3A-5.

[24] Liu W, Li M, Guzzon RS, et al. A fully reconfigurable photonic integrated signal processor. Nat Photon 2016;10:190–5.

[25] Ibrahim S, Fontaine NK, Djordjevic SS, et al. Demonstration of a fast-reconfigurable silicon CMOS optical lattice filter. Opt Express 2011;19:13245–56.

[26] Zhuang L, Zhu C, Corcoran B, et al. Sub-GHz-resolution C-band Nyquist-filtering interleaver on a high-index-contrast photonic integrated circuit. Opt Express 2016;24:5715–27.

[27] Rahim A, Schwarz S, Bruns J, et al. 16-Channel O-OFDM demul-tiplexer in silicon photonics. In: Optical Fiber Communication Conference (OFC), San Francisco, USA, 2014:Th3F-2.

[28] Zhuang L, Hoekman M, Beeker W, et al. Novel low-loss wave-guide delay lines using Vernier ring resonators for on-chip multi-λ microwave photonic signal processors. Laser Photon Rev 2013;7:994–1002.

[29] Smit MK, van Dam C. Phasar-based WDM-devices: princi-ples, design and applications. J Sel Topic Quantum Electron 1996;2:236–50.




[30] Shimizu S, Cincotti G, Wada N. Demonstration and perfor-mance investigation of all-optical OFDM systems based on arrayed waveguide gratings. Opt Express 2012;20:B525–34.

[31] Cvetojevic N, Jovanovic N, Lawrence J, et al. Developing arrayed waveguide grating spectrographs for multi-object astronomical spectroscopy. Opt Express 2012;20:2062–72.

[32] Metcalf AJ, Kim HJ, Leaird DE, et al. Integrated line-by-line opti-cal pulse shaper for high-fidelity and rapidly reconfigurable RF-filtering. Opt Express 2016;24:23925–40.

[33] Saavedra BG, Kaiser R, Beyer J, et al. 8-Channel INP OFDM transmitter PIC with integrated optical Fourier transform. In: European Conference on Optical Communication (ECOC), Düsseldorf, Germany, 2016:PW.4.P1.

[34] Goldshtein N, Sinefeld D, Golani O, et al. Fine resolution photonic spectral processor using a waveguide grating router with permanent phase trimming. J Lightwave Technol 2016;34:379–85.

[35] Feng S, Qin C, Shang K, et al. Rapidly reconfigurable high-fidelity optical arbitrary waveform generation in heterogeneous photonic integrated circuits. Opt Express 2017;25:8872–85.

[36] Sasayama K, Okuno M, Habara K. Photonic FDM multichannel selector using coherent optical transversal filter. J Lightwave Technol 1994;12:664–9.

[37] Melloni A, Morichetti F. Tunable photonic circuits: a leap toward system-on-a-chip optical integration. SPIE Newsroom, Jun. 2012. DOI: 10.1117/2.1201206.004292.

[38] Dong P, Qian W, Liang H, et al. Thermally tunable silicon racetrack resonators with ultralow tuning power. Opt Express 2010;18:20298–304.

[39] Guzzon RS, Norberg EJ, Parker JS, et al. Integrated InP-InGaAsP tunable coupled ring optical bandpass filters with zero inser-tion loss. Opt Express 2011;19:7816–26.

[40] Cincotti G. Fiber wavelet filters. J Quantum Electron 2002;38:1420–7.

[41] Hillerkuss D, Winter M, Teschke M, et al. Simple all-optical FFT scheme enabling Tbit/s real-time signal processing. Opt Express 2010;18:9324–40.

[42] Wang X, Zhou L, Li R, et al. Continuously tunable ultra-thin silicon waveguide optical delay line. Optica 2017;4:507–15.

[43] Zhang W, Yao J. Silicon-based integrated microwave photonics. J Quantum Electron 2016;52:1–12.

[44] Rasras MS, Gill DM, Patel SS, et al. Demonstration of a fourth-order pole-zero optical filter integrated using CMOS processes. J Lightwave Technol 2007;25:87–92.

[45] Takiguchi K, Oguma M, Shibata T, Takahashi H. Demultiplexer for optical orthogonal frequency-division multiplexing using an optical fast-Fourier-transform circuit. Opt Lett 2009;34:1828–30.

[46] Norberg EJ, Guzzon RS, Nicholes SC, Parker JS, Coldren LA. Programmable photonic lattice filters in InGaAsP-InP. Photon Technol Lett 2010;22:109–11.

[47] Zhuang L, Hoekman M, Oldenbeuving RM, Boller KJ, Roeloffzen CG. CRIT-alternative narrow-passband waveguide filter for microwave photonic signal processors. Photon Technol Lett 2014;26:1034–7.

[48] Corcoran B, Zhu C, Schröder J, et al. Multipass performance of a chip-enhanced WSS for Nyquist-WDM sub-band switching. J Lightwave Technol 2016;34:1824–30.

[49] Fandiño JS, Muñoz P, Doménech D, Capmany J. A monolithic integrated photonic microwave filter. Nat Photon 2017;11:124–9.

[50] Zhuang L. Flexible RF filter using a nonuniform scissor. Opt Lett 2016;41:1118–21.

[51] Schippers H, Verpoorte J, Jorna P, et al. Broadband optical beam forming for airborne phased array antenna. In: Proc. IEEE Aerospace Conference, Big Sky, MT, USA, 2009:3.0101.

[52] Soganci IM, Tanemura T, Williams KA, et al. Monolithically integrated INP 1 × 16 optical switch with wavelength-insensitive operation. Photon Technol Lett 2013;22:143–5.

[53] Tanizawa K, Suzuki K, Toyama M, et al. Ultra-compact 32 × 32 strictly-non-blocking si-wire optical switch with fan-out LGA interposer. Opt Express 2015;23:17599–606.

[54] Clements WR, Humphreys PC, Metcalf BJ, Kolthammer WS, Walsmley IA. Optimal design for universal multiport interferom-eters. Optica 2016;3:1460–5.

[55] Lu L, Zhao S, Zhou L, et al. 16 × 16 Non-blocking silicon optical switch based on electro-optic Mach-Zehnder interferometers. Opt Express 2016;24:9295–307.

[56] Suzuki K, Tanizawa K, Suda S, et al. Broadband silicon pho-tonics 8 × 8 switch based on double-Mach-Zehnder element switches. Opt Express 2017;25:7538–46.

[57] Rasras MS, Kang I, Dinu M, et al. A programmable 8-bit optical correlator filter for optical bit pattern recognition. Photon Tech-nol Lett 2008;20:694–6.

[58] Meijerink A, Roeloffzen CGH, Meijerink R, et al. Novel ring resonator-based integrated photonic beamformer for broad-band phased array receive antennas – part I: design and performance analysis. J Lightwave Technol 2010;28:3–18.

[59] Zhuang L, Roeloffzen CGH, Meijerink A, et al. Novel ring resonator-based integrated photonic beamformer for broad-band phased array receive antennas – part II: experimental prototype. J Lightwave Technol 2010;28:19–31.

[60] Roeloffzen CG, Oldenbeuving RM, Timens RB, et al. Integrated optical beamformers. In: Optical Fiber Communication Confer-ence (OFC), Los Angeles, USA, 2015:Tu3F-4.

[61] Heck MJR. Highly integrated optical phased arrays: photonic integrated circuits for optical beam shaping and beam steer-ing. Nanophotonics 2017;6;93–107.

[62] Carolan J, Harrold C, Sparrow C, et al. Universal linear optics. Science 2015;349:711–6.

[63] Harris NC, Steinbrecher GR, Prabhu M, et al. Quantum trans-port simulations in a programmable nanophotonic processor. Nat Photon 2017;11:447–52.

[64] Perez D, Gasulla I, Capmany J. Software-defined recon-figurable microwave photonics processor. Opt Express 2015;23:14640–54.

[65] Zhuang L, Roeloffzen CGH, Hoekman M, Boller KJ, Lowery AJ. Programmable photonic signal processor chip for radiofre-quency applications. Optica 2015;2:854–9.

[66] Miller DAB. Silicon photonics: meshing optics with applica-tions. Nat Photon 2017;11:403–4.

[67] Capmany J, Gasulla I, Perez D. Microwave photonics: the pro-grammable processor. Nat Photon 2016;10:6–8.

[68] Editorial. Birth of the programmable optical chip. Nat Photon 2016;10:1.

[69] Pérez D, Gasulla I, Crudgington L, et al. Multipurpose silicon photonics signal processor core. Nature Commun 2017;8:636.

[70] Brown S, Vranesic Z. Fundamentals of digital logic with Verilog design. Toronto, Canada, Tata McGraw-Hill Education, 2007.

[71] Leuthold J, Koos C, Freude W. Nonlinear silicon photonics. Nat Photonics 2010;4:535–44.




[72] Dolgaleva K, Ng WC, Qian L, Aitchison JS. Compact highly-non-linear AlGaAs waveguides for efficient wavelength conversion. Opt Express 2011;19:12440–55.

[73] Yi X, Huang TXH, Minasian RA. Tunable and reconfigurable pho-tonic signal processor with programmable all-optical complex coefficients. Trans Microw Theory Technol 2010;58:3088–93.

[74] Miller DAB. Designing linear optical components. Opt Photon-ics News 2013;24:38.

[75] Geng Z, Xie Y, Zhuang L, et al. Photonic integrated circuit imple-mentation of a sub-GHz-selectivity frequency comb filter for optical clock multiplication. Opt Express 2017;25:27635–45.

[76] Krylov AA, Sazonkin SG, Lazarev VA, et al. Ultra-short pulse generation in the hybridly mode-locked erbium-doped all-fiber ring laser with a distributed polarizer. Laser Physics Lett 2015;12:065001.

[77] Xie Y, Zhuang L, Boller K, Lowery AJ. Lossless microwave photonic delay line using a ring resonator with an integrated semiconductor optical amplifier. J Optics 2017;19;065802.

[78] Nakazawa M, Hirooka T, Ruan P, Guan P. Ultrahigh-speed ‘orthogonal’ TDM transmission with an optical Nyquist pulse train. Opt Express 2012;20:1129–40.

[79] Lowery AJ, Xie Y, Zhu C. Systems performance comparison of three all-optical generation schemes for quasi-Nyquist WDM. Opt Express 2015;23:21706–18.

[80] Fan Y, Epping JP, Oldenbeuving RM, et al. Optically integrated InP-Si3N4 hybrid laser. J Photon 2016;8:1–11.

[81] Hosseini N, Dekker R, Hoekman M, et al. Stress-optic modulator in triplex platform using a piezoelectric lead zirconate titanate (PZT) thin film. Opt Express 2015;23: 14018–26.



Documents

Yiwei Xie, Leimeng Zhuang* and Arthur J. Lowery Picosecond