ECR等离子蚀刻和湿化学蚀刻的比较

Electron Cyclotron Resonance (ECR) reactive ion etching of InP-based waveguide structures was studied using CH4/H2/Ar chemistry. The ECR process was first optimized on InP substrates before being used to process waveguides. Stripes of 2 µm width were patterned in silicon nitride and used as masking to etch strip-loaded waveguides. These waveguides were compared with wet-etched waveguides, in order to identify the dominant loss mechanism. Fabry-Perot loss measurements showed values as low as 1.1 dB/cm for the ECR-etched waveguides. From the comparison it appears that roughness of the sidewalls is more important than surface damage for the loss of these waveguides.

Indium phosphide is the material of choice for monolithic integration of optical components. This material system is suitable for application in optical communication. Photonic Integrated Circuits (PICs) on InP create the possibility for mass production to fulfill the increasing demand for more complex optical functionality at ever-lower prices.

The basic building block for PICs is the optical waveguide. Usually this is a ridge etched in a layer stack on an InP-substrate. Various etching techniques have been developed over the years. Reactive Ion Etching (RIE, [1]) has emerged as the most flexible and effective technique to obtain these waveguides. One specific problem associated with RIE is surface damage caused by impact of ions and by the chemical process. Various adjustments to RIE have been proposed and developed to reduce this problem. Electron Cyclotron Resonance (ECR) is expected to result in lower impact damage, because the plasma is ignited in the ECR column away from the sample, which is placed on the lower electrode in the chamber. We have developed an ECR-process for waveguides on InP [2]. In order to examine the loss mechanism we also realized wet-chemically etched waveguides. These have no surface damage, but are more sensitive to the sidewall roughness because of their different shape.

The waveguide is designed to provide monomodal propagation for both the TE and the TM polarization. This requirement limits the thickness of the waveguiding layer, the width of the ridges and the etching depth. The layer stack (see fig.1) is grown on an undoped InP-substrate and consists of two also undoped layers. The waveguiding layer, with a thickness of 600 nm, is an InGaAsP-layer with a bandgap of 0.956 eV, corresponding to a vacuum wavelength of 1.3 µm. The top layer of InP is 300 nm thick. Simulations with a vectorial waveguide solver [3] showed that waveguides as narrow as 1.5 µm would be multimodal for TM if the ridge is etched 100 nm into the quaternary layer. To avoid this the etching depth was limited to the thickness of the top layer: 300 nm. This results in a strip-loaded waveguide. The selective wetchemical etch will also provide this depth, making comparison of dry- and wet-chemical etched waveguides possible.

Fig1

The width of the waveguides is chosen as small as possible. This has two reasons. First of all this makes the test for the etching technique most valuable, since narrow waveguides have higher losses due to the strong interaction of the propagating field with the etched surfaces and the sidewalls. The second reason is that one of the envisioned applications is a passive polarization converter [4], which requires narrow waveguides for efficient operation. The waveguide width is chosen to be 2 μm.

From our results it is seen that the scattering from the sidewall roughness is dominant over the etching defects, since the losses from the wet-etched waveguides are appreciably higher. This in turn implies that the ECR-process used here results in a relatively defect-free etched surface. Further improvement of waveguide losses should therefore concentrate on improving the photolithography, in order to reduce the roughness of the sidewalls.