Silicon-on-insulator eight-channel optical multiplexer based on a cascade of asymmetric Mach–Zehnder interferometers

Dae Woong Kim; Assia Barkai; Richard Jones; Nomi Elek; Hat Nguyen; Ansheng Liu

doi:10.1364/OL.33.000530

Wavelength division multiplexing (WDM) is a technology of transmitting multiple optical signals of different wavelengths on a single mode fiber, which increases the transmission capacity and enhances flexibility of network configurations in optical communication systems. Recently, WDM devices based on silicon have attracted a great deal of attention, because they offer tremendous potential for low-cost and highly integrated optical components using conventional complementary metal-oxide semiconductor (CMOS) manufacturing technology [1, 2, 3]. A multiplexer (Mux) plays an important role in WDM systems. Silicon-based optical Muxes can be categorized into Mach–Zehnder interferometers (MZIs) [4] and arrayed-waveguide gratings (AWGs) [5, 6, 7]. Here, we demonstrate an eight-channel optical Mux that integrates seven asymmetric Mach–Zehnder interferometers (AMZIs) on a silicon-on-insulator (SOI) substrate. The AMZI consists of a $3 dB 1 \times 2$ splitter and a $3 dB 2 \times 2$ coupler connected to each other with different lengths of waveguides. The difference in length between the two arms dictates the period of the interferometric pattern at the output. The AMZIs can be cascaded as interleavers, increasing the number of transmitting channels. The thermo-optic effect was used to compensate phase mismatch, which can be caused by fabrication tolerances or temperature instability by using aluminum thin film heaters deposited on one arm of each AMZI; the thermo-optic coefficient for silicon is $1.86 \times 10^{- 4} K^{- 1}$ [8]. To our knowledge, this is the first report of a $1 \times 8$ $Mux$ based on cascaded AMZIs in an SOI with active tuning; while such a Mux has been reported in silica-on-silicon [9], the advantage of an SOI-based Mux is a smaller footprint.

Figure 1 shows a schematic of the eight-channel Mux consisting of three stages of AMZIs. Each AMZI is composed of a $1 \times 2$ multimode interferometer (MMI) and a $2 \times 2$ MMI. The third stage has a difference in length between the two arms of $Δ L$ , the second stage of $2 \cdot Δ L$ , and the first of $4 \cdot Δ L$ . The channel spacing $(s)$ and the difference in arm length are related with the following equation:

s = \frac{λ_{0}^{2}}{2 \cdot n_{g} \cdot (4 \cdot Δ L)},

where

λ_{0}

is the center resonance wavelength and

n_{g}

is the group refractive index of waveguides. The Mux was designed to have a channel spacing of

3.2 nm

(400 GHz)

with

4 \cdot Δ L = 97.2 μ m

\sim 1550 nm

. The length of the

1 \times 2

MMI was designed to be

54.4 μ m

; that of the

2 \times 2

MMI to be

220 μ m

; for both the widths is

7 μ m

. Figure 2 shows a cross sectional view of an SOI rib waveguide on which thin metal film (Al) heaters were deposited to form the heater. The width

(W)

, height

(H)

, and rib etch depth

(h)

of the waveguide were measured to be 0.55, 0.5, and

0.2 μ m

, respectively. The waveguide was designed to be compatible with a modulator for integration [1], and all waveguides employed a lateral taper with an

11 μ m

width at input and an

800 μ m

length for both ends to reduce coupling loss. An interlayer dielectric layer (ILD), which provides electrical insulation and planarization, was placed between the heater and waveguide to vary their separation. The cross sectional area of the Al layer is

\sim 2.5 μ m^{2}

, and the respective heater lengths for stages 1, 2, and 3 are 797.2, 608.6, and

484.3 μ m

. The footprint of the whole device

(length \times width)

is

8 mm \times 1 mm

.

A superluminescent light emitting diode (SLED) was used as a broadband light source for testing; a tapered single mode fiber with a mode field diameter of $3 μ m$ was used for coupling; and a linear polarizer and a polarization controller were interleaved between the light source and input fiber to set the desired polarization. An optical spectrum analyzer (OSA) was used to measure the output spectrum. All waveguides were antireflection coated (ARC) on both end facets to prevent undesired spectral modulations from the etalon formed between the fiber and waveguide facet.

The propagation loss of straight waveguides was measured to be $0.82 \pm 0.04 dB ∕ cm$ for TE polarization $\sim 1.55 μ m$ with the cutback method [8]. The waveguide supports only a single TE mode. The excess loss, channel spacing, and extinction ratio of a single stage $1 \times 2$ AMZI was measured to be $0.2 \pm 0.07 dB$ , $3.21 nm$ , and $28 dB$ over the wavelength from 1537 to $1562 nm$ , respectively. Figure 3a shows the phase tuning efficiencies of the heater employed in stage 1 for different ILD thicknesses as a function of applied power to the heater. The phase tuning efficiency $(η)$ can be defined by the following equation:

η = \frac{2 \cdot Δ λ}{FSR},

where

Δ λ

is the wavelength shift of resonance and FSR is the free spectral range. The FSRs for stages 1, 2, and 3 are 6.4, 12.8, and

25.6 nm

, respectively. The minima of resonance spectra were traced instead of the peak values, since they are free from distortion caused by the other two AMZIs in the same light path. Thinner ILD thickness results in better tuning efficiency, since heat is dispersed through the ILD layer. For a

1 μ m

ILD thickness,

\sim 55 mW

of power was needed to produce a π-phase shift corresponding to a

\sim 7 ° C

temperature increase for the stage 2 heated waveguide. Figure 3b shows the tuning efficiency, herein defined as the power needed for the π-phase shift for each stage. Each stage shows similar characteristics in terms of tuning efficiency for different ILD thicknesses. On average,

\sim 83 mW

of power was consumed to change the π-phase shift for each stage. In addition, by measuring the phase shift caused by adjacent heaters for all ILD thicknesses, we found that thermal cross talk was negligible for all cases (

> 6 W

was needed to produce a π-phase shift using a heater on an adjacent MZI).

Figure 4a shows the spectra of all outputs of the eight-channel Mux before the phase mismatch compensation, and Fig. 4b shows the spectra of the fully optimized Mux after heater tuning. All heaters were wire bonded to a test board [printed circuit board (PCB)] and connected to the power supply through it, which is controlled by a LABVIEW program. Applied powers to the heaters for stages 2 and 3 were 13 and $125 mW$ , respectively, and thus total applied power was $\sim 138 mW$ . The measured excess loss and the adjacent channel isolation of the optimized Mux were 2.6 and $13 dB$ , respectively. The channel uniformity, defined as the difference between the best and the worst cases in excess loss over all channels, was $\sim 1.5 dB$ over 1537 to $1562 nm$ . The loss slowly decreased as the wavelength increased owing to the wavelength dependence of the MMI, which was proved in simulation. In practice, the wavelength dependence is a key factor in determining the operating wavelength range. The channel spacing and $3 dB$ bandwidth were $\sim 3.21$ and $3 nm$ , respectively.

We successfully demonstrated an eight-channel optical Mux based on an AMZI and an SOI rib waveguide. Thermal tuning was used to compensate phase mismatch due to fabrication, and a total power of $138 mW$ was consumed to optimize all the channels. The fully tuned Mux shows a low waveguide propagation loss and high adjacent channel isolation. Furthermore, the smaller the separation between the waveguide and heater, the better the tuning efficiency, with each stage showing similar characteristics for all ILD thicknesses. In addition, thermal cross talk for the design of this device is negligible.

Fig. 1 Micrograph and schematics of an eight-channel Mux based on AMZIs; $4 \cdot Δ L = 97.2 μ m$ for $Δ λ = 3.2 nm$ $\sim 1550 nm$ , where $Δ L$ and $Δ λ$ are the difference in length between two arms of the AMZI and channel spacing $(3.2 nm)$ , respectively.

Download Full Size | PDF

Fig. 2 Cross sectional SEM pictures of an SOI rib waveguide. (a) ILD thickness= $5 μ m$ , (b) waveguide width $(W) = 0.55 μ m$ , ridge waveguide height $(H) = 0.5 μ m$ , and rib etch depth $(h) = 0.2 μ m$ .

Download Full Size | PDF

Fig. 3 Phase tuning efficiency; (a) phase shifts caused by the heater employed in stage 1 for different ILD thicknesses and (b) power consumed for the π-phase shift for each stage.

Download Full Size | PDF

Fig. 4 Spectra of all eight channel Mux outputs (a) before tuning and (b) after tuning.

Download Full Size | PDF

1. A. Liu, L. Liao, D. Rubin, H. Nguyen, B. Ciftcioglu, Y. Chetrit, N. Izhaky, and M. Paniccia, Opt. Express 15, 660 (2007). [CrossRef] [PubMed]

2. A. Liu, H. Rong, R. Jones, O. Cohen, D. Hak, and M. Paniccia, J. Lightwave Technol. 24, 1440 (2006). [CrossRef]

3. P. Dainesi, A. Küng, M. Chabloz, A. Lagos, Ph. Flückiger, A. Ionescu, P. Fazan, M. Declerq, Ph. Renaud, and Ph. Robert, IEEE Photon. Technol. Lett. 10, 660 (2000). [CrossRef]

4. Q. Lai, M. Lanker, W. Hunziker, and H. Melchior, Electron. Lett. 34, 266 (1998). [CrossRef]

5. P. D. Trinh, S. Yegnanarayanan, F. Coppinger, and B. Jalali, IEEE Photon. Technol. Lett. 9, 940 (1997). [CrossRef]

6. M. R. T. Pearson, A. Bezinger, A. Delage, J. W. Fraser, S. Janz, P. E. Jessop, and D.-X. Xu, Proc. SPIE 3953, 11 (2000). [CrossRef]

7. H. Uetsuka, IEEE J. Sel. Top. Quantum Electron. 10, 393 (2004). [CrossRef]

8. G. T. Reed and A. P. Knights, Silicon Photonics: an Introduction (Wiley, 2003), p. 92.

9. K. Tanaka, K. Iwashita, Y. Tashiro, S. Namiki, and S. Ozawa, OFC Techn. Digest 1, 88 (1999).

Silicon-on-insulator eight-channel optical multiplexer based on a cascade of asymmetric Mach–Zehnder interferometers

Abstract

Cited By

Figures (4)

Equations (2)

Optics Letters