Silicon nitride based plasmonic components for CMOS back-end-of-line integration

Silicon nitride waveguides provide low propagation loss but weak mode confinement due to the relatively small refractive index contrast between the Si3N4 core and the SiO2 cladding. On the other hand, metalinsulator-metal (MIM) plasmonic waveguides offer strong mode confinement but large propagation loss. In this work, MIM-like plasmonic waveguides and passive devices based on horizontal Cu-Si3N4-Cu or CuSiO2-Si3N4-SiO2-Cu structures are integrated in the conventional Si3N4 waveguide circuits using standard CMOS backend processes, and are characterized at 1550-nm telecom wavelengths. The Cu-Si3N4(∼100 nm)-Cu devices exhibits ∼0.8-dB/μm propagation loss for straight waveguides, ∼38% coupling efficiency with the conventional 1-μm-wide Si3N4 waveguide through a 2-μm-long taper coupler, ∼0.2-dB bending loss for sharp 90° bends, and ∼0.1-dB excess loss for ultracompact 1×2 and 1×4 power splitters. Inserting a ∼10-nm SiO2 layer between the Si3N4 core and the Cu cover (i.e., for the corresponding Cu-SiO2-Si3N4-SiO2-Cu devices), the propagation loss and the coupling efficiency are improved to ∼0.37 dB/μm and ∼52% while the bending loss and the excess loss are degraded to ∼3.2 dB and ∼2.1 dB respectively due to the step index modulation. Ultracompact plasmonic ring resonators with 1-μm radius are demonstrated with exhibit extinction ratio of ∼18 dB and quality factor of ∼84, close to the theoretical prediction. ©2013 Optical Society of America OCIS codes: (240.6680) Surface plasmons; (250.5403) Plasmonics; (130.3120) Integrated optics devices; (230.4170) Multi-layers. References and links 1. A. Biberman, K. Preston, G. Hendry, N. Sherwood-Droz, J. Chan, J. S. Levy, M. Lipson, and K. Bergman, “Photonic network-on-chip architectures using multilayer deposited silicon materials for high-performance chip multiprocessors,” ACM J. Emerging Technologies in Computing Systems 7(2), article 7 (2011). 2. N. Schewood-Droz and M. Lipson, “Scalable 3D dense integration of photonics on bulk silicon,” Optics Express 19(18), 17758-17765 (2011). 3. A. Gondarenko, J. S. Levy, and M. Lipson, “High confinement micron-scale silicon nitride high Q ring resonator,” Optics Express 17(14), 11366-11370 (2009). 4. L. Razzari, D. Duchesne, M. Ferrera, R. Morandotti, S. Chu, B. E. Little, and D. J. Moss, “CMOScompatible integrated optical hyper-parametric oscillator,” Nature Photonics 4, 41-45 (2010). 5. R. Halir, Y. Okawachi, J. S. Levy, M. A. Foster, M. Lipson, and A. L. Gaeta, “Ultrabroadband supercontinuum generation in a CMOS-compatible platform,” Optics Lett. 37(10), 1685-1697 (2012). 6. D. K. Gramotnev and S. I. Bozhevolnyi, “Plasmonics beyond the diffraction limit,” Nat. Photon. 4, 83-91 (2010). 7. R. F. Oulton, G. Bartal, D. F. P. Pile, and X. Zhang, “Confinement and propagation characteristics of subwavelength plasmonic modes,” New J. of Phys. 10, art. 105018 (2008). 8. L. Chen, J. Shakya, and M. Lipson, “Subwavelength confinement in an integrated metal slot waveguide on silicon,” Optics Lett. 31(14), 2133-2135 (2006). 9. S. Zhu, T. Y. Liow, G. Q. Lo, and D. L. Kwong, “Silicon-based horizontal nanoplasmonic slot waveguides for on-chip integration,” Optics Express 19(9), 8888-8902 (2011). 10. S. Y. Zhu, G. Q. Lo, and D. L. Kwong, “Components for silicon plasmonic nanocircuits based on Cu-SiO2Si-SiO2-Cu nanoplasmonic waveguides,” Optics Express 20(6), 5867-5881 (2012). 11. S. Zhu, G. Q. Lo, and D. L. Kwong, “Phase modulation in horizontal metal-insulator-silicon-insulatormetal plasmonic waveguides,” Optics Express 21(7), 8320-8330 (2013). 12. S. Y. Zhu, G. Q. Lo, and D. L. Kwong, “Experimental demonstration of vertical Cu-SiO2-Si hybrid plasmonic waveguide components on an SOI platform,” IEEE Photon. Techn. Lett. 24(14), 1224-1226 (2012). 13. H. S. Lee, C. Awada, S. Boutami, F. Charra, L. Douillard, and R. E. de Lamaestre, “Loss mechanisms of surface plasmon polaritons propagation on a smooth polycrystalline Cu surface,” Optics Express 20(8), 8974-8981 (2012). 14. N. N. Feng, M. L. Brongersma, and L. D. Negro, “Metal-dielectric slot-waveguide structures for the propagation of surface plasmon polaritons at 1.55 μm,” IEEE J. Quantum Electronics 43(6), 479-485 (2007). 15. X. T. Kong, W. G. Yan, Z. B. Li, and J. G. Tian, “Optical properties of metal-multi-insulator-metal plasmonic waveguides,” Optics Express 20(12), 12133-12146 (2012). 16. W. Cai, W. Shin, S. Fan, and M. L. Brongersma, “Elements for plasmonic nanocircuits with threedimensional slot waveguides,” Adv. Mater. 22(45), 5120-5124 (2010). 17. Z. Han, “Ultracompact plasmonic racetrack resonators in metal-insulator-metal waveguides,” Photonics and Nanostructures Fundamentals and Applications 8, 172-176 (2010). 18. P. Neutens, P. Van Dorpe, I. D. Vlaminck, L. Lagae, and G. Borghs, “Electrical detection of confined gap plasmons in metal-insulator-metal waveguides,” Nat. Photonics 3, 283-286 (2009). 19. P. Neutens, L. Lagae, G. Borghs, and P. Van Dorpe, “Plasmon filters and resonators in metal-insulatormetal waveguides,” Optics Express 20(4), 3408-3423 (2012). 20. L. A. Sweatlock and K. Diest, “Vanadium dioxide based plasmonic modulators,” Optics Express 20(8), 8700-8709 (2012). 21. E. Feigenbaum, K. Diest, and H. A. Atwater, “Unity-order index change in transparent conducting oxides at visible frequencies,” Nano Lett. 10, 2111-2116 (2010). 22. http://www.lumerical.com. 23. S. Roberts, “Optical properties of copper,” Phys. Rev. 118, 1509-1518 (1960). 24. S. Y. Zhu, G. Q. Lo, and D. L. Kwong, “Experimental demonstration of integrated horizontal Cu-Si3N4-Cu plasmonic waveguide and passive components,” Photonic Global Conf (PGC), Singapore, Dec. 13 (2012). 25. J. Tian, S. Yu, W. Yan, and M. Qiu, “Broadband high-efficiency surface-plasmon-polariton coupler with silicon-metal interface”, Appl. Phys. Lett. 95, art. 013504 (2009). 26. Z. Han, V. Van, W. N. Herman, and P. T. Ho, “Aperture-coupled MIM plasmonic ring resonators with subdiffraction modal volumes,” Opt. Express, 17(15), 12678-12684 (2009). 27. S. Y. Zhu, G. Q. Lo, and D. L. Kwong, “Experimental demonstration of horizontal nanoplasmonic slot waveguide-ring resonators with submicrometer radius,” IEEE Photon. Techn. Lett. 23(4), 1896-1898 (2011).


Introduction
Silicon nitride (Si 3 N 4 ) is transparent in both the visible and infrared spectrum and can be deposited on almost any substrate using mature technologies such as low pressure chemical vapor deposition (LPCVD) and plasma enhanced chemical vapor deposition (PECVD), thus it has become one of the promising materials for integrated photonics applications, especially for three-dimensional (3D) integration of multiple photonics above the processed microelectronics [1,2].Propagation loss as low as 0.1 dB/cm and intrinsic quality factor of ring resonators as high as 3×10 6 have been demonstrated on the Si 3 N 4 platform [3].Moreover, active functions such as parametric amplification [4] and broadband supercontiniuum generation [5] have been realized using the nonlinear properties of Si 3 N 4 .However, due to the relatively small refractive index contrast between the Si 3 N 4 core and the oxide cladding (Δn ∼ 0.5), the optical mode in Si 3 N 4 waveguides is usually loosely confined and the minimum allowable bend radius is relatively large (e.g., >20 μm [3]), which results in large footprint for Si 3 N 4 -based photonic devices and low integration density for Si 3 N 4 -based photonic circuits.
On the other hand, deep subwavelength plasmonic waveguides such as metal-insulatormetal (MIM) can confine the optical mode at nanometer scale and allow sharp bending, thus enabling miniaturization of the photonic devices and dense integration [6][7][8].However, the high confinement plasmonic waveguides suffers from large propagation loss due to the unavoidable metal loss [7].A straightforward solution for this issue is to integrate the high confinement plasmonic waveguides and the conventional low-loss dielectric waveguides in the same chip, in which the plasmonic waveguides are used to realize ultracompact photonic devices while the conventional low-loss dielectric waveguides are used to transmit optical signals over long distances [8].For cost efficiency and ease of fabrication with electronics, this photonic and plasmonic integrated circuit should still be CMOS compatible.In silicon-oninsulator (SOI) platform, CMOS-compatible photonic and plasmonic integrated circuits have been demonstrated based on horizontal Cu-insulator-Si-insulator-Cu nanoplasmonic waveguides [9][10][11] or vertical Cu-insulator-Si hybrid plasmonic waveguides [12].Copper is the metal of choice because it is widely used in CMOS backend processes and it offers lower metal loss at 1550-nm wavelengths than aluminum [13].
Here, CMOS-compatible plasmonics for integration in Si 3 N 4 -based photonic circuits is addressed.The plasmonic waveguide has horizontal Cu-Si 3 N 4 -Cu or Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu structure.These two structures can be catalogued in general as the MIM waveguide.To date, MIM waveguides have been thoroughly investigated and many MIM-based photonic devices have been theoretically proposed [14][15][16][17].However, only a few of them have been demonstrated experimentally [18,19], probably because a typical MIM waveguide, which has a layered metal-dielectric-metal structure with Au or Ag as the metal, is not CMOS compatible and is difficult to connect with the low-loss dielectric waveguide.The horizontal MIM-like waveguides developed in this paper provide an alternative way to realize various proposed MIM plasmonic devices cost-effectively.

Experimental
Fabrication was started from 200 mm silicon wafers.∼4-μm SiO 2 was deposited by PECVD, followed by chemical mechanical polishing (CMP) to smooth the surface (Fig. 1(a)).Then, ∼375-nm Si 3 N 4 was deposited by LPCVD.The deposition was carried out for two times (first ∼200 nm, then ∼175 nm) to prevent stress-induced cracking (Fig. 1(b)).After patterning using UV lithography, Si 3 N 4 was dry etched to leave ∼49-nm Si 3 N 4 slab layer using photoresistor as the mask (Fig. 1(c)).After removing the photoresistor, a thick SiO 2 layer was deposited by PECVD (Fig. 1(d)), followed by SiO 2 window opening by dry etching SiO 2 down to the Si 3 N 4 slab layer (Fig. 1(e)).A thin dielectric layer can be deposited as an optional step to finally form the so-called metal-multi-insulator-metal plasmonic waveguides [14].Interesting functionalities such as electro-optic modulation may be introduced if a unique dielectric such as VO 2 [20] or ITO [21] is deposited.Here, for simplification, a thin SiO 2 layer is deposited on some wafers by PECVD (Fig. 1(f)).Then, 150-nm Cu was deposited by sputtering, followed by 1-μm Cu electroplating (Fig. 1(g)).After low-temperature annealing to improve the Cu film quality [13], Cu-CMP was carried out to remove Cu (as well as the deposited thin SiO 2 layer) outside the windows (Fig. 1 Figure 2(a) shows microscope picture of one of the fabricated photonic devices.The Cucovered plasmonic area is inserted in the conventional Si 3 N 4 waveguides through linear taper couplers, as shown schematically in Fig. 2(b).Figure 2(c) is the cross-sectional transmission electron microscopy (XTEM) image of one of the fabricated plasmonic waveguides.A Si 3 N 4 core with ∼49-nm slab is covered by ∼10-nm SiO 2 almost conformally.Due to the imperfection of fabrication, the cross section of the Si 3 N 4 core is not an ideal rectangle, but is triangle-shaped with sidewall angle of ∼83° at the bottom region and ∼65° at the top region.The Si 3 N 4 core shown in Fig. 2(c) has bottom width of ∼90 nm and height (etch depth) of ∼288 nm above the ∼49-nm Si 3 N 4 slab.In fabrication, the bottom width of the Si 3 N 4 core is tuned from ∼110 nm to ∼80 nm by changing the exposure dose during the UV lithography.Due to the angled sidewall, the Si 3 N 4 core with the bottom width larger than ∼110 nm is trapezoid-shaped with height of ∼326 nm while that with the bottom width smaller than ∼110 nm is triangle-shaped with a reduced height.Plasmonic waveguides without the thin SiO 2 interlayer (i.e., Cu-Si 3 N 4 -Cu waveguides) were also fabricated.The diced chips were measured using a fiber-to-fiber measurement setup.Quasi-TEpolarized light (the electric field is parallel to the chip surface plane) is coupled into the input Si 3 N 4 waveguide from a lensed polarization-maintaining (PM) single-mode fiber, transports through the waveguide, and then is coupled out to another fiber to be measured by a power meter and an optical spectrum analyzer (OSA).A semi-auto micrometer piezo-stage was used to adjust both input and output fibers to search the maximum output power.
Numerical simulation was performed using commercial software Lumerical [22].The refractive indices of Si 3 N 4 and SiO 2 are set to 2.0 and 1.444, respectively, and the complex index of Cu at 1550 nm is set to 0.181 + j11.05 [13,23].

Si 3 N 4 rib waveguides
The Si 3 N 4 waveguides (fabricated along with the plasmonic devices in the same chip) are characterized first.Figure 3(a) shows that it has a rib structure with ∼892-nm width, ∼375-nm height, and ∼326-nm etch depth, embedded in the thick SiO 2 cladding layer.The electrical field (|E x |) distribution of the fundamental 1550-nm TE mode is depicted in Fig. 3(b), calculated using the eigen-mode expansion (EME) method [22].The calculated effective modal index is 1.626 and the ratio of optical power contained in the Si 3 N 4 core is 60% (i.e., the remaining 40% optical power is contained in the cladding SiO 2 layer).The propagation loss measured by the standard cutback method is plotted in Fig. 3(c) as a function of wavelength ranging from visible to infrared, using different laser sources which are available in our laboratory.The propagation loss is smaller than ∼1 dB/cm over a wide wavelength range, comparable to those reported in literature for single-mode Si 3 N 4 channel waveguides [2,3].The loss increase in the C-band and the large loss (∼10.5 dB/cm) at 420 nm may be attributed to Si-H and N-H bonds in the film.Waveguide ring resonators (WRRs) with radii ranging from 20 μm to 50 μm were fabricated.They exhibit typical resonant characteristics with extinction ratio of ∼20-dB, loaded Q-value of ∼1100, and group index of ∼1.9 [24] (not shown here).It indicates that the subsequent processes for Cu-based plasmonic devices do not degrade the processed Si 3 N 4 waveguides in the same chip.

Plasmonic waveguides
Figure 4(a) plots spectra measured on a set of straight Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguides with the same Si 3 N 4 core but different lengths (L P s) using a broadband (1520-1620 nm) laser source, normalized by that measured on a reference Si 3 N 4 waveguide without the plasmonic area.The small ripples in the spectra may arise from the weak reflection at the taper couplers.Except these ripples, the spectra are almost wavelength-independent in this wavelength range.The output power depends on L P almost linearly, as shown in Fig. 4(b), from which the propagation loss is extracted by linear fitting: ∼0.46 dB/μm for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide and ∼0.64 dB/μm for the Cu-Si 3 N 4 -Cu waveguide.The waveguides were also measured using other laser sources.Unlike the conventional Si 3 N 4 waveguides, the plasmonic waveguides exhibit larger propagation loss at shorter wavelengths, for example, the propagation loss of the above Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide is ∼0.88 dB/μm at 1310 nm, ∼3.2 dB/μm at 1060 nm, and >5 dB/μm at 853 nm.The larger propagation loss at the shorter wavelength can be attributed to the larger metal loss of Cu at the shorter wavelength [13,23].Since low propagation loss is essential for all plasmonic devices, only the broadband 1550-nm laser source is used in the following measurement.The dependence of propagation loss and real part of the effective modal index (n eff ) on the size of the Si 3 N 4 core is shown in Figs. 5 (e) and 5(f) for both Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu and Cu-Si 3 N 4 -Cu waveguides.The experimental data (represented by the symbols) are measured using the cutback method and the theoretical data (represented by the solid curves) are calculated using the EME method.For comparison, the plasmonic waveguides with 326-nmhigh ideal rectangular Si 3 N 4 cores are also calculated (represented by the dash curves).Firstly, for all kinds of waveguides, the propagation loss increases with the core size decreasing, reflecting the fact that the tighter mode confinement leads to the larger propagation loss.Secondly, the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguides have lower propagation loss and smaller n eff than the corresponding Cu-Si 3 N 4 -Cu waveguides evening scaled to the same total size of the core.This observation can be explained by the fact that the latter confines the mode more tightly than the former.Thirdly, the experimental propagation loss of Cu-Si 3 N 4 -Cu waveguides is larger than the theoretical value, which may be attributed to the dry-etching induced sidewall roughness of the Si 3 N 4 core.The roughness induced scattering loss may be suppressed by the thin SiO 2 layer deposited above the Si 3 N 4 core.As a result, the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguides have similar experimental propagation loss as the theoretical value.Fourthly, the real imperfect Si 3 N 4 core differs in waveguide properties from the ideal rectangular Si 3 N 4 core in the range of bottom width smaller than ∼110 nm.However, this apparent difference can be simply attributed to the reduction of the height of the real triangleshaped Si 3 N 4 core.In the range of bottom width larger than ∼110 nm, the waveguides with the trapezoid core and the ideal rectangular core exhibit similar properties if the width of the real trapezoid core is scaled to the effective width, which is approximate to the middle width of the trapezoid core.It indicates that the angled sidewall does not affect the waveguide's properties significantly.Therefore, the Si 3 N 4 core is approximate to a 326 nm (height) × 100 nm (width) rectangle in the following numerical simulations for simplification.

Taper couplers
As the abovementioned, an effective coupling between the plasmonic waveguide and the lowloss dielectric waveguide is essential for the photonic and plasmonic integrated circuits.Here, a linear taper coupler as shown schematically in Fig. 2 Plasmonic waveguides with different coupler length (L C ) (with the same L P of 15 μm) were fabricated.Figure 6(c) plots output spectra measured on Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguides with L C of 0, 0.3, 1, and 5 μm, normalized by that of the reference Si 3 N 4 waveguide without the plasmonic area.The spectra depend on wavelength weakly except the ripples.Figure 6(d) plots the coupling loss versus L C for Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu and Cu-Si 3 N 4 -Cu waveguides.For both waveguides, the coupling loss decreases with L C increases quickly, reaches a minimum at L C of 1-2 μm, and then increases slowly with L C further increasing.This is because the coupling loss includes the tapering loss (due to reflection and radiation) and the propagation loss through the taper [8], the tapering loss decreases while the propagation loss increases with L C increasing.The optimal taper length is ∼1-2 μm for both waveguides with minimum coupling loss of ∼2.8 dB/facet for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide and ∼4.2 dB/facet for the Cu-Si 3 N 4 -Cu waveguide.The coupling loss for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide is close to the theoretical value while that for the Cu-Si 3 N 4 -Cu waveguide is larger than the theoretical value, which can also be attributed to the sidewall roughness induced scattering loss.For comparison, a similar taper coupler between the Si channel waveguide and Cu-SiO 2 -Si-SiO 2 -Cu plasmonic waveguide can provide a coupling loss as low as ∼0.1 dB/facet [8,9,25].The relatively large coupling loss for the Si 3 N 4 -based waveguides in this work may be attributed to the relatively small mode overlap between the Si 3 N 4 waveguide and the plasmonic waveguide because of the loose mode confinement in the Si 3 N 4 waveguide.From Figs. 6(a) and 6(b), it indicates that the mode power contained in the Si 3 N 4 core will be tapered into the Cu-covered coupler while that contained in the SiO 2 cladding will be scattered, which results in ∼2.2 dB tapering loss as only 60% mode power confined in the Si 3 N 4 core.The n eff mismatch between the Si 3 N 4 waveguide and the plasmonic waveguide is another source of tapering loss, while it should be small for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide because its n eff is close to that of the Si 3 N 4 waveguide, as observed from Fig. 5(f).The Cu-Si 3 N 4 -Cu waveguides have even smaller mode size and larger n eff than the corresponding Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide, thus they have larger tapering loss, as observed in Fig. 6(d).

Ultracompact 90° bends
It has been theoretically verified that the high confinement MIM plasmonic waveguide allows sharp bending [16].In this work, 90° bends with radius (R) of 0 and 0.5 μm were fabricated, as shown in Figs.7(a) and 7(c).The total length of total plasmonic route is 3 μm and 3.8 μm, respectively.The top view of field distributions in these two bends, obtained from 3D FDTD simulation are shown in Figs.7(b) and 7(d) respectively.The theoretical bending loss is near to zero.Experimentally, the bending loss is extracted by comparing the output power measured on the bend with that measured on the corresponding 3-μm-long plasmonic waveguide.The measurement results are plotted in Fig. 7(e).One sees that the spectra are almost wavelength-independent except the ripples.For the Cu-Si 3 N 4 -Cu plasmonic waveguides, the average bending loss is ∼0.2 dB for R ∼ 0 μm and ∼0.7 dB for R = 0.5 μm.The slightly large bending loss of the 0.5-μm-R bend is due to its longer plasmonic route (3.8 μm) than the reference straight plasmonic waveguide (3.0 μm).Therefore, the experimental result agrees well with the theoretical prediction.On the other hand, the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguides exhibit relatively large bending loss.The bending loss is ∼3.2 dB for R ∼ 0 μm and is ∼1.9 dB for R = 0.5 μm, which may be attributed to its large mode size as compared to the corresponding Cu-Si 3 N 4 -Cu waveguide.

Ultracompact 1×2 and 1×4 power splitters
It has been theoretically verified that the MIM plasmonic waveguide supports ultracompact power splitters [16].In this work, ultracompact 1×2 splitters and 1×4 splitters were fabricated, as shown in Figs.8(a) and 8(d) respectively.The opening angle is 90° and the length of each plasmonic route is 3 μm.The 3D FDTD simulation shows that these splitters can split the input optical power to each output port equally with near zero excess loss, as shown in Figs.8(b) and 8(e), respectively.Experimentally, each output port of these splitters is measured and compared with that of the corresponding 3-μm-long straight plasmonic waveguide.Figure 8(c) and 8(f) plots the measured spectra for 1×2 and 1×4 splitters, respectively.One sees that the spectra are almost wavelength-independent except small ripples.For the 1×2 splitters, the input light is split almost equally to each output port.The average excess loss is ∼0.1 dB for the Cu-Si 3 N 4 -Cu splitters and ∼2.1 dB for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu splitters.The large excess loss for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu 1×2 splitters can also be attributed to its large mode size as compared to the corresponding Cu-Si 3 N 4 -Cu 1×2 splitters.For the 1×4 splitters, the power delivered to the out-1 and out-4 ports is slightly smaller than that delivered to the out-2 and out-3 ports, which may be attributed to the imperfection of the fabricated junctions [10].For the Cu-Si 3 N 4 -Cu splitters, the average output power delivered to the out-1 and out-4 ports is ∼ -7.2 dB and that delivered to the out-2 and out-3 ports is ∼ -6.2 dB, which corresponds to an average excess loss of ∼0.7 dB and non-uniformity of ∼1.0 dB.For the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu splitters, the average output power delivered to the out-1 and out-4 ports is ∼ -7.7 dB and that delivered to the out-2 and out-3 ports is ∼ -6.2 dB, which corresponds to an average excess loss of ∼1.0 dB and non-uniformity of ∼1.5 dB.Again, the large excess loss (as well as the large non-uniformity) for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu 1×4 splitters can be attributed to its large mode size as compared to the corresponding Cu-Si 3 N 4 -Cu 1×4 splitters.

Plasmonic waveguide ring resonators
Ultracompact MIM plasmonic waveguide ring resonators (WRRs) have been theoretically investigated [17].For evanescently coupling between the bus and ring MIM plasmonic waveguides, the gap between them should be smaller than ∼26 nm because the penetration depth of optical filed in the metal is around 26 nm.Such a small gap cannot be defined using the conventional UV lithography, but requires an expensive electron beam lithography or focused ion beam lithography.To circumvent this problem, an aperture coupler has been proposed theoretically for MIM WRRs [26] and has been demonstrated experimentally for horizontal Cu-SiO 2 -Si-SiO 2 -Cu plasmonic WRRs [27].In this work, ultracompact WRRs based on the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide are fabricated, as shown in Fig. 9(a).The bus waveguide has L P = 7 μm and L C =2 μm, and has the same cross sectional structure as the ring waveguide.The outer radius of the ring is 1 μm.The final dimension of the aperture coupler is determined by the nominal gap size in the mask (which ranges from 0 to 0.2 μm) as well as the exposure dose during the UV lithography [27], which is very sensitive to the performance of the plasmonic WRRs.WRRs with optimal aperture size exhibit typical resonant properties.The experimental spectrum agrees well with the theoretical spectrum obtained from the 3D FDTD simulation, which is given as the dash curve in Fig. 9(b).The electric field distributions in the ring at the off-resonance state and on-resonance state are shown in Figs.9(c) and 9(d), respectively, and the dynamic 1550-nm TE light propagation through this plasmonic WRR is shown by the attached movie (Media 1).One sees that light goes through the bus waveguide at the off-resonance state and is trapped in the ring at the on-resonance state, similar to the conventional WRRs.The Cu-Si 3 N 4 -Cu WRRs also exhibit resonance property but with smaller extinction ratio and smaller quality factor (not shown here) due to its relatively large propagation loss as compared with the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguides.

Conclusion
Si 3 N 4 -based plasmonic waveguides and various passive devices including bends, power splitters, and ring resonators have been experimentally demonstrated for seamlessly integrating them in the Si 3 N 4 -based photonic circuits.A thin SiO 2 interlayer between the Si 3 N 4 core and the Cu cover can modify the devices' properties significantly.Relatively low propagation loss of ∼0.37 dB/μm and coupling loss of ∼2.8 dB/facet have been measured on horizontal Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguides and ultracompact ring resonators with radius of 1 μm have been demonstrated with extinction ratio of ∼18 dB and quality factor of ∼84 around telecom 1550-nm wavelengths.On the other hand, the horizontal Cu-Si 3 N 4 -Cu waveguides supports sharp bending and splitting with near-zero loss.Processes and materials used here are compatible and common to the standard CMOS backend processes, thus it opens the door to integrate dense photonic and plasmonic integrated circuits on the backend of CMOS electronics.

Fig. 2 .
Fig. 2. (a) Microscope picture of one of the fabricated devices; (b) Schematic layout of the horizontal Cu-dielectric-Si 3 N 4 -dielectric-Cu waveguide inserted in the conventional Si 3 N 4 waveguide through taper couplers with length of L C ; and (c) XTEM image of one of the fabricated plasmonic waveguides.

Fig. 3 .
Fig. 3. (a) XTEM image of the fabricated Si 3 N 4 rib waveguide; (2) Electric field |E x | distribution of the fundamental 1550-nm TE mode in the waveguide calculated using EME method; and (c) Propagation loss versus wavelength, measured using different laser sources.

Fig. 4 .
Fig. 4. (a) Output spectra measured on a set of straight Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguides with different L P s, normalized by that measured on the reference Si 3 N 4 waveguide without the plasmonic area; and (2) The measured output power (normalized by the reference waveguide) versus L P and the linear fittings.

Figures 5 (
Figures 5(a)-5(d) show electric field (|E x |) and the magnetic field (|H y |) distributions of the 1550-nm fundamental TE mode in the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu and Cu-Si 3 N 4 -Cu waveguides, calculated using the EME method.One sees that the optical mode is tightly confined in the core for both waveguides.For the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide, the electric field in the thin SiO 2 interlayer is significantly enhanced due to the continuity of electric displacement normal to the Cu/SiO 2 and SiO 2 /Si 3 N 4 interfaces.The ratio of modal power in the core is ∼87% for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide and ∼90% for the Cu-Si 3 N 4 -Cu waveguide.It indicates that the Cu-Si 3 N 4 -Cu waveguide provides tighter mode confinement (i.e., smaller mode size) than the corresponding Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide.

Fig. 5 .
Fig. 5. (a) The electric field (|E x |) and (b) the magnetic field (|H y |) distribution of the 1550-nm fundamental TE mode in the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguide, (c) |E x | and (d) |H y | distribution in the Cu-Si 3 N 4 -Cu plasmonic waveguide, (e) Propagation loss and (f) Real part of the modal index (n eff ) versus the bottom width of the Si 3 N 4 core for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu and Cu-Si 3 N 4 -Cu plasmonic waveguides.
(b) is used, which is validated theoretically by 3D finite-difference time-domain (FDTD) simulation.The electric field distributions depicted in Figs.6(a) and 6(b) shows that the 1550-nm fundamental TE mode in the 1-μm-wide Si 3 N 4 waveguide is effectively fanned in the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguide through a 2-μm-long taper coupler.The theoretical coupling loss is ∼2.9 dB for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguide and ∼3.3 dB for the Cu-Si 3 N 4 -Cu waveguide.Experimentally, the coupling loss can be extracted from the y-intercept of linear fitting lines shown in Fig. 4(b), by assuming that the in-coupling and the out-coupling have the same coupling loss.The coupling loss extracted from Fig. 4(b) is ∼3.2 dB/facet for the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu waveguide and ∼4.2 dB/facet for the Cu-Si 3 N 4 -Cu waveguide.

Fig. 6 .
Fig. 6.(a) Top view and (b) Cross sectional view of the electric field distribution of the 1550nm TE mode coupling from the 1-μm-wide Si 3 N 4 waveguide to the Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguide through a 2-μm-long taper coupler.(c) Output spectra measured on Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu plasmonic waveguides with L P = 15 nm and L C ranging from 0 to 5 μm, normalized by that measured on a reference Si 3 N 4 waveguide without the plasmonic area.(d) The measured coupling loss versus L C for Cu-SiO 2 -Si 3 N 4 -SiO 2 -Cu and Cu-Si 3 N 4 -Cu plasmonic waveguides.

Fig. 7 .
Fig. 7. (a) SEM image of Si 3 N 4 core of a 90° bend with R ∼ 0 μm, (b) The electric field distribution in a Cu-Si 3 N 4 -Cu 90° bend with R = 0, obtained from the 3D FDTD simulation; (c), (d) corresponding figures for a 90° bend with R = 0.5 μm, (e) Experimental bending loss spectra, obtained by subtracting the spectrum measured on the corresponding 3-μm-long straight plasmonic waveguide from those measured on the 90° bands.

Fig. 9 .
Fig. 9. (a) SEM image of the Si 3 N 4 core of a plasmonic waveguide ring resonator with radius of 1 μm, (b) Theoretical (obtained from 3D FDTD simulation) and experimental transmission spectra of a plasmonic WRR, normalized by that of the corresponding 7-μm-long straight plasmonic waveguide, (c) Electric field distribution in the off-state, and (d) Electric field distribution in the on-state.

Figure 9 (
Figure 9(b) plots (the solid curve) a transmission spectrum in the range of 1520-1620 nm measured on a plasmonic WRR with normal gap of 0.15 μm, normalized by that measured on a corresponding 7-μm-long straight plasmonic waveguide.It exhibits one resonance at λ r = 1564 nm with extinction ratio of ∼18 dB, insertion loss of ∼2 dB, and quality factor is ∼84.The experimental spectrum agrees well with the theoretical spectrum obtained from the 3D FDTD simulation, which is given as the dash curve in Fig. 9(b).The electric field