Flat-top interleavers:A novel approach based on MMI splitters

In most telecom applications, optical band-pass filters must ensure flat-top response. Different filter synthesis techniques have been proposed in the literature, including lattice filters [1] and cascaded interferometers with suitably chosen symmetries [2,3]. All these methods require in general the possibility to choose freely the splitting ratios of the power splitters, which (at least in theory) is usually not too hard to achieve with directional couplers. On the other hand, Multimode Interferometers (MMIs) are an attractive alternative to directional couplers, thanks to their more relaxed tolerances to fabrication errors and their relatively broadband operation. Nevertheless, they have the drawback that only a limited discrete set of five splitting ratios is achievable with a single MMI, namely 50:50, 85:15, 73:27, and their inverse ratios [4]. This issue clearly limits their use in designing flat-top interferometric filters. In this work we show for the first time that is possible to design 4-stage flat-top interferometers using four 50:50 MMIs (i.e. two double MMIs [5]) and three 85:15 MMI splitters. Like in Ref. [3], the design approach is based on the representation of the system on the Bloch sphere for two-level systems. Flat-top interleavers with different free spectral ranges have been designed and fabricated on the standard micron-scale silicon photonics platform of VTT, based on 3 µm thick rib and strip waveguides. Two main types of layouts have been explored: a more straightforward one where all components are collinear as shown in Fig. 1(a), and a more compact one shown in Fig.1(b), where the components have been folded in a spiral shape. All interleavers have been designed for TE polarization, and they work fine in a wavelength range comparable with the 100 nm bandwidth of the MMI splitters. Most designs show in-band extinction exceeding 15 dB. Fabrication imperfections and non-ideal behaviour of both bends and MMIs led to reduced extinction compared to simulations and, interestingly, also to clear differences in the spectra corresponding to different input ports, as clearly shown in Fig. 1(c). The in-band losses of the most central channels did not exceed 1.5 dB compared to the fiber-to-fiber losses measured with the reference straight waveguide